Early Season Inefficiencies in the NFL Sports Betting Market
______
A Thesis
Presented to
The Honors Tutorial College
Ohio University
______
In Partial Fulfillment
of the Requirements for Graduation
from the Honors Tutorial College
with the degree of
Bachelor of Business Administration
______
by
Michael D. DiFilippo
June 2012
Early Season Inefficiencies in the NFL Sports Betting Market 2
This thesis has been approved by
The Honors Tutorial College and the Department of Finance
Dr. Andrew Fodor Assistant Professor, Finance Thesis Advisor
Dr. Raymond Frost Director of Studies, Business Honors Tutorial College
Dr. Jeremy Webster Dean, Honors Tutorial College
Early Season Inefficiencies in the NFL Sports Betting Market 3
Reflection
For the two articles discussed in this reflection, there were four authors including myself. The authors of the paper were: Dr. Andy Fodor, Ohio University;
Dr. Kevin Krieger, University of West Florida; Dr. Justin Davis, Ohio University and myself. All of the aforementioned contributed to both papers and as such have been listed as authors. The two papers began as a single paper. After completing the project as a single paper we determined that the paper would have better publication prospects if it were divided into two papers. Below is a detailed description of how these two papers came to be, including a reflection on what I learned through the process of writing the papers.
Generation of Topic
The topic for the paper was formulated by Dr. Andy Fodor and Dr. Kevin
Krieger. Both Dr. Fodor and Dr. Krieger have written extensively on the topic of sports betting market efficiency and therefore are knowledgeable about the sports betting market literature. The topic of the paper was formed when Dr. Fodor and Dr.
Krieger surmised that betting markets would be inefficient during the first week of the
NFL season due to a lack of current season information. Thus, bettors would increasingly rely on outdated previous season information when making bets.
The professors figured that given the high amount of yearly turnover in NFL team rosters, the performance of teams season-over-season could vary substantially.
Therefore, if betting lines were influenced by last season’s performance, they are likely to be inaccurate relative to games later in the season as current season Early Season Inefficiencies in the NFL Sports Betting Market 4
information is available. They contended that the inaccuracy in betting lines would allow for the forming of a profitable betting strategy that exploits this first week inefficiency in the NFL sports betting market.
It was shortly after Drs. Fodor and Krieger decided to pursue this topic as a potential paper, that I began my tutorial with Dr. Fodor. During the first week of my tutorial, Dr. Fodor and I talked about this new idea and he suggested that we focus my tutorial around working on this paper. He explained that working on the paper would be a great learning experience for me because it would give me the opportunity to complete an empirical analysis with a large data set and also help me further understand the process of writing a scholarly article.
We concluded that I would do all of the work on the paper including the empirical analysis, literature review and writing of the paper. Every week during the quarter I was given a task to complete and have ready for discussion and review during the following week’s tutorial. Our plan was to work on the empirical analysis first, complete the written sections of the paper (data and methodology, results and conclusion) and then revise and edit during the winter break.
Empirical Analysis of Data
Before I could get started with the empirical analysis, Dr. Krieger had to gather the data for the paper. In order to ensure that we had a sufficient sample size to perform our empirical analysis, Dr. Fodor and Dr. Krieger decided to test the hypothesis over 1,917 games played during the 2004-2011 NFL regular seasons. Dr.
Krieger collected all of the sports betting lines and game results from Early Season Inefficiencies in the NFL Sports Betting Market 5
sportsinsight.com, which tracks spreads and results for numerous sports. After collecting and organizing the data, Dr. Krieger passed the data along to Dr. Fodor and
I and I began to work on the empirical analysis.
Going into the empirical analysis I was very worried that I was going to struggle because I had not taken a statistics class since the winter quarter of my freshman year. However, Dr. Fodor did a great job of explaining the different methods to be used and why these methods were appropriate. The key to my success with the empirical analysis was the strong foundation Dr. Fodor set before we even began doing the analysis. He had a clear plan for the tests that needed to be performed to test our hypothesis.
The first step in performing the empirical analysis was to determine whether the inefficiency that Drs. Fodor and Krieger hypothesized existed at a statistically significant level. Before, I was able to go about documenting the inefficiency, I first had to go through and categorize the games by the number of prior-season playoff teams that were playing in the game. We distinguished teams as prior-season playoff teams if that team had played in the playoffs in the previous season. Therefore, each game was given an identifier of 0, 1 or 2. We surmised that if a team made the playoffs during the previous season (signifying strong performance in that season) betting lines would be biased in their favor (making them too big a favorite or too small an underdog) in the first week of the next season when playing a team that did not make the playoffs in the prior season. Early Season Inefficiencies in the NFL Sports Betting Market 6
After breaking down the games by the number of prior-season playoff teams playing in the game, I began to look for the inefficiency that Drs. Fodor and Krieger had hypothesized. This process involved calculating line errors from each game when a prior playoff team played a prior non-playoff team during week one of the eight seasons. I then calculated mean and median line errors and tested for statistical differences from zero.
The average line error is calculated as the score of the playoff team (adjusted by the point spread) less the score of the non-playoff team. If the error is positive, the playoff team won by more than expected. If the line error is negative it means the playoff team did not beat the spread and therefore, the betting line was too high. If the line error is zero betting line was exactly correct. If the markets were truly efficient, one would expect that the mean (median) error would not differ significantly from 0, signifying that the markets have accurately predicted the outcome of the games.
However, if the average line error is statistically different from 0, this is evidence of a bias in spreads signifying that an inefficiency exists.
Whenever I conducted the above test, I found that the average line error for week 1 games when a prior playoff team played a prior non-playoff team was -4.28, which is significantly different from 0 at the 5% level. Comparatively, I calculated the line error for weeks 2-17 and found an average line error of 0.58, which is not statistically different than 0. The error of -4.28 in the first week of the playoffs showed that the betting lines during week 1 of the NFL season were too high for prior playoff Early Season Inefficiencies in the NFL Sports Betting Market 7
teams when playing prior non-playoff teams. This was evidence that the inefficiency
Drs. Fodor and Krieger hypothesized did truly exist in the NFL sports betting market.
The average line error that was found in Week 1 of NFL seasons is very significant to the literature because few prior papers identify inefficiencies that are significant across multiple seasons. After finding the hypothesized inefficiency, I began the process of further substantiating the finding, and exploring the implications of the finding, by conducting various other empirical analyses. .
Some of the other analyses I conducted included exploring historical playoff appearance trends, prior playoff team betting lines and win percentages, performance of prior playoff teams against the spread, as well as some testing on over/under betting. Throughout the other analyses I conducted, I found many interesting results pertaining to our paper. One of the interesting findings is that on average only half of the teams that play in the playoffs in any particular season will return to the playoffs in the next season. This was particularly interesting because under our hypothesis, bettors in the first week see past performance as a predictor for future performance. However, according to that statistic, prior performance is not a good indicator of future performance, since only half of playoff teams return to the playoffs in the next season.
Another interesting finding I made through my empirical analysis was that because of the bias bettors have for these prior playoff teams in the first week of the season, only 34% of the prior playoff teams beat the spread in the first week when playing a prior non-playoff team. We found that 34% is statistically different from
47.62% at the 10% level. 47.62% is used because it is necessary to win 52.38% of Early Season Inefficiencies in the NFL Sports Betting Market 8
games to make a profit when taking into account commissions charged by bookmakers. The test performed is equivalent to assuming the bettor bets again prior playoff teams (winning 66% of bets) and testing for a statistical difference from
52.38%.This low percentage of winning against the spread supports the presence of a bias towards prior playoff teams documented by the line error results.
One of the most intriguing findings of the empirical analysis was evidence of a
“learning effect”. Our results showed that the bias we had identified in the first week of the NFL season dissipates quickly as the season progresses. We attribute this finding to bettors relying less on past season information as information on the current season’s teams becomes available.
The most important finding that came out of the empirical analysis was from tests for profitability of a betting strategy based on exploiting the inefficiency we document. One of the weaknesses in many of the previous articles written on the efficiency of the NFL sports betting market was that bettors couldn’t profitably exploit the inefficiencies that were identified. Before starting the write up phase of the testing, we decided it was important to check to see if we could have made a profit exploiting the inefficiency that we discovered. Given that prior playoff teams had a significantly smaller chance of beating the spread in a game, we decided the best strategy to exploit this inefficiency would be to bet against all of the prior playoff teams in the first week of the NFL seasons when these teams were matched against prior non-playoff teams.
When we checked the profitability of the betting strategy across the 2004-2011 seasons, we were amazed at what we found. The profitability of our simply strategy Early Season Inefficiencies in the NFL Sports Betting Market 9
was 25.6% per game! A profit margin this high is unheard of in the sports betting literature.
Writing of the Rough Draft
After the completion of our empirical analysis, Dr. Fodor and I worked together to double check the statistical findings and then began the process of starting the rough draft of the paper. The first aspect of writing the rough draft was for me to write the descriptions of all of the tables we had created while doing the empirical analysis. This was an important step in the process of writing the paper because it allowed me to take the time to figure out what exactly the results in each table meant and how they were going to fit into “the story” we were trying to tell in the paper. By taking the time to write the descriptions of the tables first, I was able to get a great picture of the results of our tests before I wrote the full article about our findings.
Once the table descriptions were written, the next step was for me to write a review of the existing academic literature pertaining to inefficiencies in NFL sports betting markets. I went about the process of writing the literature review by first reading all of the relevant academic journal articles that I could find and then briefly summarizing each. After reading all of the articles I was able to get a sense of which articles were relevant to our paper and which were not. Once I had selected the papers to include in my literature review, I began putting together the summaries of these papers in order to tell the story of the literature’s stance on efficiency of the NFL sports betting market. Early Season Inefficiencies in the NFL Sports Betting Market 10
Overall I found the process of writing the literature review to be fairly easy as I had completed literature reviews in the past for my previous tutorials. The most difficult part of writing the literature review was understanding terminology that is unique to the sports betting markets and statistical tests that the various papers used. I found that the easiest way to approach writing a literature review is to find the most recent substantial paper published in the respective field and then use that paper as a springboard to find relevant articles written previously.
Upon completion of the rough draft of the literature review, I began writing the other parts of the paper including the data and methodology, results and conclusion sections. The hardest part of writing these various different parts was finding the best way to convey the message in a straightforward and uncomplicated way. I found it difficult for me to take all of the knowledge that I had accrued through the empirical analysis and translate that knowledge into simple and succinct writing. The section that I had the most trouble with was the data and methodology section because it was very much focused on heavy statistical techniques and theories that I did not have a very strong grasp on at the time.
Similar to the data and methodology section, I also found the results section to be somewhat difficult to write because it was hard to determine the best order for presented the tables I had created through the empirical analysis. Once I determined the best order to present the results, I found it was fairly straightforward to tell the story of what the results of our tests showed. However, similar to the data and Early Season Inefficiencies in the NFL Sports Betting Market 11
methodology section, I did struggle with describing some of our findings due to my inexperience pertaining to writing about statistical findings for an academic journal.
After I had written the data and methodology and results sections, the only other section that I had left to write was the conclusion of the paper. I realized the conclusion section was one of the most important because it tied together everything that had previously been discussed in the paper and outlined why the article our significant to the literature. The hardest part of writing the conclusion for me was making it clear how our paper made a significant contribution to the literature. Being that I was not completely familiar with the literature, I found it difficult to properly place our article in the overall sports betting market efficiency literature. However, with guidance from Dr. Fodor, we were able to finish the conclusion section and the rough draft of the paper.
Once we had the rough draft finished, the last task that I had to complete was to write the abstract for the paper. When Dr. Fodor first told me that I had to write an abstract, I figured it would be a simple task since it is typically only a paragraph long.
However, I came to realize that the small size of the abstract makes it quite difficult to write because you are forced to describe a 20-25 page paper in only one paragraph.
After writing a couple of drafts and getting some advice from Dr. Fodor, I was able to come up with a strong abstract for the paper.
Preparing the Paper for Publication
After the rough draft had been finished, Dr. Fodor and I spent the remainder of the quarter editing and revising the paper. The majority of the edits that we did Early Season Inefficiencies in the NFL Sports Betting Market 12
together on the paper pertained to flow of sentences and word choices. One of the biggest lessons I learned from writing this paper was the importance, and difficulty, of writing about complicated subjects in a simple and straightforward way.
Also as we were editing the paper, I came to realize that my writing tended to be quite wordy and long-winded. Many of the edits that Dr. Fodor and I did focused on eliminating unnecessary words and simplifying sentences and paragraphs. There were countless times that Dr. Fodor and I were going over my paper and he would take two sentences and split them up or change their order to clear up the flow of the paragraph. One of the other keys lessons I learned while preparing the paper for submission to an academic journal was to focus on keeping my writing as short and concise as possible while still conveying the intended message.
In addition to cleaning up some of the bad habits in my writing, Dr. Fodor also spent a lot of time placing particular words and phrases into the paper that are unique to academic journal articles. Given that this was the first time I had ever written a scholarly article, I was not familiar with, many of the terms that were part of the standard vernacular for journal articles. As I went through the editing process, I was able to pick up on them and utilize them more effectively when I was editing parts of the paper. After we had spent some time cleaning up the paper, Dr. Fodor took the paper and sent it off to our co-authors, Dr. Davis and Dr. Krieger for further edits and revisions.
Over the course of winter break, Drs. Davis, Fodor and Krieger, worked on further editing and revising paper, eventually splitting it up into two different papers. Early Season Inefficiencies in the NFL Sports Betting Market 13
The two papers that came out of the original rough draft that I had written during my tutorial are entitled “Inefficient Pricing from Holdover Bias in NFL Point Spread
Markets” and “Early Season NFL Over/Under Bias”, both of which have been submitted to academic journals as part of this thesis.
Over/Under Paper
The second paper that was spun out of the original rough draft pertained to the over/under aspect of the NFL sports betting market. As part of the empirical analysis that I did with Dr. Fodor, we analyzed an additional inefficiency that existed in NFL sports betting markets in the first week of the season. Our findings showed that the totals in the over/under market gave predictions for scoring that were too high in the first week of the NFL regular season. The hypothesis was that since offenses are much more reliant on execution and timing in order to work effectively as a team, they are at a disadvantage early in the season compared to defenses which rely more heavily on player instincts.
Given that offenses are at a disadvantage in the early season due to a lack of experience in playing with each other, we hypothesized that offenses will score fewer points on average in the first week of the season relative to later weeks in the season.
Therefore, final score will be less likely to exceed the total line. The betting lines for the over/under bets are simply a prediction of how many total points will be scored in any particular game. If the two teams’ combined total score is higher than the betting line, and a bettor bets “over” then the bettor wins. Conversely, if the two teams score Early Season Inefficiencies in the NFL Sports Betting Market 14
fewer combined points than the total line, a bettor who bet “under” would win. If the bettor chooses incorrectly the bet is lost.
We hypothesized that over/under lines in the first week of the NFL season are too high, meaning games are more likely to have total score under the total line in the first week of the season than in later weeks. Therefore, we purposed a strategy of betting “under” on every game in the first week of the NFL season. After testing this betting strategy across the 8 seasons in our dataset, we found average profits of of
13.6% per game. The profits were again statistically significant, a rarity in sports betting literature.
Submission for Publication
After both of the papers had been revised and edited by my three faculty co- authors, Dr. Fodor submitted the paper entitled “Inefficient Pricing from Holdover
Bias in NFL Point Spread Markets” to The Journal of Political Economy, a top-tier journal, and the paper entitled “Early Season NFL Over/Under Bias” to The Journal of
Sports Economics, a second tier journal. Given our papers provide significant contributions to the literature due to high statistical significance and novelty and intuitiveness of our hypotheses, we are confident the papers will be published in strong, peer reviewed journals s in the near future.
Final Reflections
The process of writing my thesis taught me many invaluable lessons that will stick with me throughout the rest of my life. The most notable of these lessons are learning how to persevere when you feel overwhelmed, relying on others that are Early Season Inefficiencies in the NFL Sports Betting Market 15
smarter than you when you don’t know how to proceed and utilizing strict time management when you have a large project with a distant deadline.
The first lesson I learned through the completion of my thesis is how to persevere through tough material. Before I started the process of writing the holdover bias paper I had no exposure to sports betting literature and didn’t know the first thing about betting on sports. Needless to say, when Dr. Fodor suggested we write a paper on sports betting, I got concerned because it was a topic I knew nothing about.
However, after hearing his idea and seeing how excited he was about the paper, I became determined to write the paper and stretch my intellectual capacity. Due to my lack of knowledge about sports betting, writing the paper took extra time as I often had to look up unfamiliar topics I would come across in the literature. Slowly, I was able to pick up on the important terminology and get a strong enough understanding of the sports betting markets to write the paper. Looking back on the process of writing the paper, I have a strong sense of accomplishment given that I was able to study and write a paper on an academic topic I was not familiar with.
One of the key aspects of my success in writing the paper with little previous knowledge of the subject matter was relying on Dr. Fodor to guide and mentor me through the process. It is because of the critical role that Dr. Fodor played in the development and process of writing the paper that I was able to learn the second major lesson through this process, rely on others smarter than you whenever you don’t know how to proceed. Writing this paper was the first time I had ever written a scholarly paper, let alone a scholarly paper based on heavy empirical analysis. Initially, I was Early Season Inefficiencies in the NFL Sports Betting Market 16
very overwhelmed by the amount of work ahead of me and felt lost in the fact that I had never written a scholarly paper before. However, I voiced my concerns to Dr.
Fodor and he was able to help me develop a plan and guide me every step of the way through the writing process. Had Dr. Fodor not spent the time to mentor me and guide me through the process, I would never have been able to successful complete the paper. Working with Dr. Fodor on this paper taught me the key lesson of whenever you are lost and don’t know how to proceed, rely on those around you that are smarter and more experienced, and they will help you along the way.
The third lesson pertains to time management when you have a big project with a distant deadline. I learned this lesson both through the writing of the original paper in the fall of 2011 and also in the spring of 2012 while writing this literature review and reflection. The lesson I learned is when you have a big project with multiple steps and a distant deadline, the best thing you can do is plan out the steps and schedule a date by which you will complete each step. By having a schedule and sticking to it, you will be able to ensure that you have ample time to complete each step and you will also have a barometer in which to measure your progress along the way. For both aspects of my thesis, I came up with a schedule and an order in which to complete each step. By sticking to the schedule, I was able to reduce my stress and also maximize my effectiveness. From now on as I move into the working world, I will keep this lesson in mind and will always strive to manage my time better by creating a schedule for my long-term projects and developing checkpoints to ensure I am staying on schedule. Early Season Inefficiencies in the NFL Sports Betting Market 17
Writing this thesis has taught me a lot of valuable life lessons, only a few of which have been outlined in this reflection. The lessons that were described above, including the importance of persevering through tough work, relying on others for help and strict time management, all were crucial to my success in writing this thesis.
These lessons will also be crucial in the next stage of my life as I go out into the business world.
As I write this conclusion to my reflection on my thesis work, I can’t help but also reflect on my entire collegiate career. Looking back on the fall quarter of my freshman year and the person that I was then, there is no doubt in my mind that the freshman version of me would not have been able to write the papers that are included in this thesis. A lack of maturity, dedication and focus would have limited my ability to write a paper of this magnitude.
However, over the past three years, thanks to countless people along the way, I have matured and grown academically as well as professionally and personally. And I as sit here and put the finishing touches on my collegiate career through the writing of the last few words of this thesis, it is amazing to think back and see how far I have come. Throughout the past 3 years in college I have gone from an immature, undedicated freshman to a mature and focused business professional. This transformation could not have been possible without the many academic and life lessons I have learned along the way, a few of which I have outlined in this reflection.
As I conclude these final reflections, I want to take the opportunity to say thanks to all that have helped me throughout my academic career. The countless lessons that I have Early Season Inefficiencies in the NFL Sports Betting Market 18
learned at Ohio University will continue to resonate with me forever. Thank you to all for making me a better person and for helping to shape that immature, unfocused freshman into the mature and focused business professional that I am today.
Early Season Inefficiencies in the NFL Sports Betting Market 19
Inefficient Pricing from Holdover Bias in NFL Point Spread Markets
Andy Fodor* Ohio University Finance Department [email protected]
Michael DiFilippo Ohio University [email protected]
Kevin Krieger University of West Florida Department of Accounting and Finance [email protected]
Justin Davis Ohio University Department of Management [email protected]
Keywords: Sports Wagering, Efficient Markets, Holdover Bias, Irrational Investment
*Contact Author: 234 Copeland Hall Athens, OH 45701 (740) 593-0514
Early Season Inefficiencies in the NFL Sports Betting Market 20
Abstract
We identify inefficiency in the NFL gambling market indicative of sticky preferences by bettors. NFL teams that qualified for the playoffs in the prior season are favored by too large a margin in the opening week of the following season. Systematic betting based on this trend results in significant profitability over the 2004-2011 seasons with an average return over 25% per game. We posit this can be explained by gamblers’ tendencies to cling to perceptions of teams formed from observation in the prior season. This confirms research in more traditional markets suggesting investors can be slow to update asset valuations.
Early Season Inefficiencies in the NFL Sports Betting Market 21
Introduction
Basic efficient market theory suggests that gambling spreads serve as unbiased prices for wagers on athletic contests. The best available information regarding a game should be reflected in a spread so that each side of a handicapped contest is equally likely to prevail. 1 Bookmakers are thus able to approximately equalize the funds wagered on either side of a contest and thereby guarantee a riskless profit via commissions. This is the traditional model of bookmaking advanced in the literature
(see, e.g., Zuber et al., 1985 and Sauer et al., 1988).
The efficiency of the National Football League (NFL) betting market has been a topic of investigation in the finance literature for over forty years. Given that this market has grown into a multi-billion dollar arena for gambling, the primary stream of research has been aimed at identifying inefficiencies that are exploitable for financial gain. To this point, findings from these efforts have been mixed, at best. Numerous authors have identified trends in the betting markets that, if exploited over certain periods, would have led to significant profits. However, other authors have often found such results fail to persist out of sample. Some of the findings of systematic profitability generate from strategies lacking a strong underlying theory, thus exacerbating data mining concerns.
1 Bookmakers set point spreads, or “lines”, in the most common form of handicapping games. The point spread issued by the bookmaker (often a casino or internet company) establishes the “favorite” and the “underdog” of a game. This point spread serves as a correction based on the perceived likelihood of each team winning a game. The favorite is considered more likely to win a game, and thus, the spread is instituted in order to place the two sides of a wager on more equivalent footing. A wager is graded based on subtracting the spread from the favorite’s final score and comparing this adjusted figure to the score of the underdog. Whichever side then has the higher score is the winning team of the “against the spread” wager. The team that wins an against-the-spread wager is said to have “covered” the game or the spread. Early Season Inefficiencies in the NFL Sports Betting Market 22
Pankoff (1968) provided the initial study of efficiency in the NFL betting market, finding that the market was, overall, efficient. However, the first study considering abnormal profitability based on specific strategic wagering in the NFL was undertaken by Sturgeon (1974), who found that betting against the previous week’s biggest winner was profitable. In subsequent years numerous studies have been conducted which consider various betting strategies in search of similar profitability.
Many of these studies report success in this endeavor. Vergin and Scriabin
(1978) consider a number of betting rules of thumb and report that betting on large underdogs (those of five or more points) generated approximately a 5% profit over a five-year period. Zuber et al. (1985) find NFL games (from the 1983 season) to be profitably predictable based on a number of team measures such as rushing yards, passing yards, fumbles, and interceptions, though they do not find statistical significance, in part due to the small sample size available. Gandar et al. (1988) reconsider the betting rules of Vergin and Scriabin (1978) and introduce new rules of thumb to study. Similar to Zuber et al. (1985) they document profitability based on trading rules, but no statistical significance. Golec and Tamarkin (1991) describe biases in NFL lines against underdogs potentially leading to a profitable betting opportunity, though the economic impact is not great. More recently, Paul and
Weinbach (2011) find betting against big favorites to be profitable and statistically significant. This is confirmed independently by Wever and Aadland (2012) who note Early Season Inefficiencies in the NFL Sports Betting Market 23
that wagering on large underdogs from the 1985 through 2010 NFL seasons would have proven to be a profitable approach.
On the other hand, the documentation of profitability from such
“inefficiencies” has been called into serious question. Sauer et al. (1986) describe how wagering based on the findings of Zuber et al. (1985) leads to substantial losses in out of sample testing. Dare and Holland (2004) find that previously documented inefficiencies of profitable betting on home underdogs are not consistent from season to season. Gray and Gray (1997) find in-sample profitability based on trading rules but acknowledge, in line with the findings of Sauer et al. (1986), that results are considerably mixed out of sample. Vergin (1998) finds that profitable trading rules developed by Lacey (1990), based on the 1984-1986 NFL seasons, did not hold for the subsequent 1987-1995 period.
The difficulty in confirming the persistence, out of sample, of the profitability of betting approaches should not be unexpected. As Burkey (2005) notes, authors who search for profitable trading rules which, ex post, prove to have been successful in earlier periods, will surely succeed if they investigate enough strategies. In extreme cases, dozens of betting rules may be investigated in one study. For example,
Woodland and Woodland (2000) reject 7 of 48 null hypotheses based on a 10% significance level. Badarinathi and Kochman (1996) consider 116 null hypotheses and reject 7 based on a 5% significance level. An inference of profitability of a specific strategy, based on such a widespread approach, would very likely be committing a Early Season Inefficiencies in the NFL Sports Betting Market 24
type I error. Findings of profitability are particularly tenuous when little theoretical development is undertaken in developing a hypothesized strategy.
We consider the question of whether bettors in gambling markets display irrationally sticky preferences for wagers in a manner similar to investors who are unwilling to update their asset preferences to reflect new information. Brown and Cliff
(2005) document that investors are willing to pay unusually high premia for assets they hold in high sentiment. These high prices result in subsequent abnormally low returns, even in the presence of controls. Haruvy et al. (2007) document that individuals’ beliefs about prices adapt over time, but they are based in part on past experiences. As a more specific example, numerous authors document the willingness of investors to pay high prices for Internet stocks in the midst of the tech bubble due to sentiment (see, e.g., Cooper et al., 2002).
This framework motivates our study of the holdover bias of NFL bettors from one season to the next. Specifically, we consider the case of the early weeks of
American professional football seasons. Teams that qualified for the NFL playoffs in the prior season are systematically favored by too many points in Week 1 of the next season when they play opponents who did not qualify for last season’s playoffs (or, in the rare cases where they are underdogs, they are underdogs by too few points).
Simply wagering, against the spread, on every Week 1 opponent of last season’s playoff teams would result in an average return of 25.6% per game over the 2004-
2011 NFL seasons. Unlike many other studies of profitability from betting strategies, our results are statistically significant as well. Early Season Inefficiencies in the NFL Sports Betting Market 25
We posit that the more detailed explanation of these findings lies in the inability of the participants of this market, the bettors, to adjust their preferences. This theory is in line with the findings of Brown and Cliff (2005) and Haruvy et al. (2007) for more traditional markets. Many gamblers have a perception of a club based on the previous season. When a playoff participant faces a Week 1 opponent that did not qualify for last season’s playoffs, a substantial number of bettors are apt to frame the contest as a match-up between a “successful” and an “unsuccessful” team and wish to wager accordingly. Bookmakers, cognizant of this irrational preference by gamblers, may adjust the lines they set on a game, forgoing the traditional riskless profit model in search of an expected higher profit.2 If spreads of such games are, indeed, set in order to entice bettors to cater to their preferences at an effectively higher price, wagering against last season’s playoff participants in early weeks of NFL seasons should prove profitable.
In related work, Sapra (2008) documents that teams that perform well (poorly) in one season against the spread tend to revert in performance against the spread the following year. This is described as “overreaction” by bettors to previous impressions.
Our findings are perhaps most similar in vein to those of Vergin (2001) who documents that NFL gamblers develop preferences for teams that perform well in the previous game, two to five games, or season. He also characterizes this effect as
“overreaction” by bettors from the 1981-1995 seasons. Our results provide out of sample verification of such bettor behavior for the 2004-2011 period. However, our
2 Theoretical development and empirical validation of this modified, profit maximization framework was first undertaken by Levitt (2004) and confirmed by Paul and Weinbach (2007) and by Krieger et al. (2011) in out of sample findings. We elaborate further on this framework in the discussion section. Early Season Inefficiencies in the NFL Sports Betting Market 26
results are also different in some important ways. Unlike Sapra (2008) or Vergin
(2001), we focus on the specific case of holdover preferences from one season to the next and study how these preferences are specifically manifested in a season’s opening week. 3 In doing so, we document strong profitability levels and, in addition, our results are statistically significant, a threshold few other studies of gambling profitability meet. We focus on only one hypothesis in our formulation of this paper, unlike studies that consider multiple betting rules of thumb, and we test our results out of sample for robustness.
Our Week 1 results may emerge because longer periods of time work to more forcefully instill investor (or bettor) preferences. NFL gamblers typically adjust their impressions of teams on a weekly basis throughout a season. The final impression left at the end of one regular season, however, has over eight months to linger in the minds of bettors before gambling on Week 1 of the following season commences. The psychology literature notes that “attitude strength”, analogous here to an investor preference, is greater when an opinion has been held for a longer time (see, e.g.,
Holland et al., 2003; Glasman and Albarracin, 2006). Savvy bookmakers may exploit these preferences in search of higher expected profits, and thus, bettors willing to take
‘against the herd’ positions may profit. Bettors update their evaluations of teams in the opening weeks of a season, and thus the opportunity to profit based on sticky beliefs regarding last season dissipates after Week 1.
3 Vergin (2001) analyzes a trading rule (one of 11 he considers) which bets against all playoff teams in the following season but does not consider the holdover bias question specifically for the early portion of next season. His result regarding last year’s playoff teams is that they are no more or less likely to cover spreads in the following season. Early Season Inefficiencies in the NFL Sports Betting Market 27
The 25.6% return demonstrated by our study is unprecedented in the NFL wagering literature, even compared to strategies developed in papers, which consider dozens of potential betting rules. Vergin and Scriabin (1978) report that betting on large underdogs (those of five or more points) generated approximately a 5% profit over their five-year study period. Zuber et al. (1985) win 59% of wagers in the latter half of NFL seasons based on models built from team performance and a number of variables measured in the first half of seasons. This translates to a 12.3% return.
Gandar et al. (1988) analyze four mechanical-based and three behavior-based rules in an effort to profit in the NFL betting market. The most profitable of their strategies involves wagering on underdogs playing a favorite who easily covered the spread in the previous week and translates to an 11.1% return. Golec and Tamarkin (1991) find that betting home underdogs over the 1973-1987 period returned 6.1%. Gray and Gray
(1997) utilize a probit model to choose which teams to wager on in NFL games. Using the highest cutoff probability of winning as a threshold for wagering on a team, they demonstrate 7.7% returns in sample and (surprisingly) 16.1% returns out of sample.
Vergin (2001) analyzes 70 potential NFL betting strategies and of these 70 strategies, the most profitable returns 18.7%. Using data from 1985-2008, Wever and Aadland
(2012) find that betting on large home underdogs and even larger road underdogs yields an 11.2% return.
While the magnitude of returns we report for the Week 1, prior playoff approach is newfound, the strategy’s effectiveness is actually more impressive when considered in full context. We consider only one hypothesis in our study, and this Early Season Inefficiencies in the NFL Sports Betting Market 28
hypothesis is soundly developed theoretically, in contrast to efforts which briefly introduce numerous potential betting rules. The returns of the approach are statistically significant, a threshold that almost no other paper considering the NFL gambling market meets. The returns of the approach are not statistically significant when investigated, out of sample, via seasons in the distant past; however, the returns of the strategy are still positive in these earlier seasons, even after factoring in bookmaker commission, a standard few other studies reach out of sample. Furthermore, we detail explanations for why the approach, though recently successful, may not have been as relevant in the long-ago seasons used for out of sample analysis. Overall, the effectiveness, development and robustness of the approach are a significant contribution to the literature.
The rest of this study is presented as follows. First, we provide theoretical support for our suggested hypothesis. Next, we describe our data collection methods and the statistical methodology used. We then present results of the study. For robustness, we describe the performance of our strategy out of sample. In doing so we also discuss reasons why profitability of the strategy may continue going forward. We conclude in the final section.
Hypothesis Development
In the traditional model of bookmaking, bookmakers attempt to set a line for a game so that equal amounts of money will be wagered on each side, and the Early Season Inefficiencies in the NFL Sports Betting Market 29
bookmaker may then claim a riskless profit due to the 10% commission that is charged on winning bets (e.g. Zuber et al., 1985; Sauer et al., 1988).4 A refined model has been developed in recent years, claiming that bookmakers are willing to instead take some level of risk in order to increase their expected profitability. Bookmakers may do so by setting lines so that naïve bettors will flock to the side of a contest which is less than
50% likely to prevail in a wager.
Avery and Chevalier (1999) note that a great deal of “dumb money” exists in the betting market and that bettors make errors in their wagering based on preferences for teams, visibility of teams, and momentum beliefs. A bookmaker might ignore this fact and set a line that achieves equal betting on each side of a game, or the bookmaker may shift the line in order to intentionally allow more of the betting to occur on the side of a contest less likely to win. Thus, by assuming some risk on individual contests, the bookmaker may increase profits over time. We offer the following example for clarification. A perfectly knowledgeable bookmaker handicaps the actual spread of a game between the New Orleans Saints and Cleveland Browns to be the Saints -9 points. The perfect bookmaker also knows most bettors would prefer to bet on the Saints (or. perhaps, against the Browns) and they are so eager to do so that betting would not be equalized between the two sides unless the line were Saints -
11 points. The bookmaker may set the line at Saints -10 points and therefore attract more than 50% of the money wagered on the Saints and be more than 50% likely to win these bets. This approach would result in an expected profitability greater than the
4 As another example, Lee and Smith (2002) note, “Bookies do not want their profits to depend on the outcome of the game. Their objective is to set the point spread to equalize the number of dollars wagered on each team...” Early Season Inefficiencies in the NFL Sports Betting Market 30
5% guaranteed from the riskless, commission-only approach. Given a large enough bankroll, over time, a large number of games handicapped thusly will minimize any risk to the bookmaker.5
The profit maximization approach is discussed by Strumpf (2003) who notes that hometown bookies may set spreads, which local teams are notably less than 50% likely to cover. Levitt (2004) conducted a seminal study confirming this approach by bookmakers in the NFL, based on special data from an NFL handicapping contest at the Las Vegas Hilton, and this finding was later supported by Paul and Weinbach
(2007) and Krieger et al. (2011).
One might ask if informed bettors (often professional gamblers known as
“sharps” or “wiseguys”) might not take advantage of such lines and nullify the bookmaker’s profitability by wagering on the “correct” side of a contest. While sharps undoubtedly seek out such opportunities, their impact is limited by three factors. First, a sharp cannot be certain that his handicapping of any one game is superior to the bookmaker, and thus, risk aversion dictates prudence. Second, the bankroll of a sharp is typically considerably less than that of the bookmaker, and here again, risk aversion prevents sharps from too heavily exploiting even a spread they confidently identify as biased. Third, the impact a sharp can make in a betting market like the NFL is
5 While this structure supposes a “perfect” bookmaker, anecdotal documentation exists that real bookmakers are, indeed, willing to take such an approach to improve profits (see Millman, 2001). Early Season Inefficiencies in the NFL Sports Betting Market 31
saturated by ordinary, naïve bettors. Betting limits placed on gamblers by internet companies or casinos further limit this impact.6
Given the documentation of this framework, we posit that bookmakers intentionally set spreads in such a manner as to attract naïve bettors to wager on teams for which they’re biased. We further describe below why this bias is likely to be based on sentiment for teams perceived as “successful” as demonstrated by their post-season presence in the previous NFL year. Bettors willing to take the contrarian approach may thus be able (like the sharps) to place wagers with positive expected value, even factoring in bookmaker commissions.
Barberis et al. (1998) build a model describing how investor sentiment may impact asset returns. They cite the conservatism bias noted by Edwards (1968) in the psychology literature that holds that people are slow to update beliefs based on information. When beliefs are altered, the updates typically undershoot the true value that should be reflected. In our study we hypothesize a direct parallel for participants in the betting market for NFL games. In particular, we believe that bettors are cognizant that the end of one NFL season marks an important shift in the quality of a team. Substantial personnel turnover of both players and coaches occurs in between seasons, and many teams, particularly unsuccessful ones from the previous year, are likely to make dramatic strategic shifts in the eight-month offseason. Nevertheless, many investors may hold too tightly to their perceptions of success from the previous season and be willing to wager accordingly. Thus, the prices offered by bookmakers,
6 Typical maximum bets range between $10,000 and $25,000 for one side of an NFL contest. Any wagers larger than the maximum amount must be broken into separate wagers, allowing the bookmaker to change the spread in the interim. Early Season Inefficiencies in the NFL Sports Betting Market 32
in the form of lines, may be set at specific levels in order to take advantage of these biases.
Broad empirical evidence exists supporting Barberis et al.’s (1998) theory that investor sentiment inflates asset prices and therefore yields lower future returns. Neal and Wheatley (1998) consider three different measures of investor sentiment: odd-lot sales frequencies, closed-end fund discounts, and net mutual fund redemptions. They note that the first two proxies help predict stock returns. Fisher and Statman (2000), rather than studying different proxies of sentiment, consider three different groups of stock market participants and conclude that the preferences of all three are negatively linked to future returns. Wall Street strategists and individual investor sentiment are each significantly negatively linked to the performance of stocks in the future while the sentiment of newsletter writers is insignificant but still negative in direction.
Brown and Cliff (2005) note that it is difficult to demonstrate inefficiency due to sentiment in traditional markets. They conduct a survey of investors and note new evidence of inefficiency due to sentiment. Excessive optimism of investors is linked to future periods of market-wide overvaluation. The result is economically significant as well. In a laboratory setting, Haruvy et al. (2007) show that investors update prices over time, but prices are based on past trends that traders have experienced in markets.
This is analogous to our theory that bettors hold on to perceptions from the prior NFL season. While prices converge to fundamentals in Haruvy et al.’s experiment, bubbles do persist for a time. Early Season Inefficiencies in the NFL Sports Betting Market 33
The evidence supporting the negative link between investor sentiment and asset returns is not restricted to domestic stocks. Nor is the impact of sentiment demonstrated only by identifying which stocks are held in highest esteem and separating those stocks for comparison. Schmeling (2009) empirically confirms the sentiment theory in the international marketplace by studying 18 industrialized countries and noting that, across nations, investor sentiment is negatively linked to future stock returns. The findings are particularly strong in those countries more prone to overreaction. The empirical findings of the impact of sentiment are not restricted to stock markets either. Han (2008) demonstrates that beyond factors typically modeled, index option prices reflect the sentiment surrounding the equity market. We suggest that this investor sentiment argument is also applicable in the sports betting framework.
Data and Methodology
Our data covers 1,917 games played in the 2004-2011 NFL regular seasons.
All sports betting lines and game results are collected from sportsinsights.com, which tracks spreads and results for numerous sports.7
We hypothesize that inefficiencies may exist in gambling markets in the early weeks of an NFL season due to bettors overemphasizing team performance in the prior year. We choose to explore the implications of prior season playoff appearances (and thus utilize the terminology “prior playoff teams”) because bettors might classify a team as having a successful year when it qualifies for the postseason. After one NFL
7 Line data from sportsinsights.com reflects historical closing lines from pinnaclesports.com. Early Season Inefficiencies in the NFL Sports Betting Market 34
season ends, this perception of success (or failure, in the case of non-playoff teams) could remain the impression of a club, for bettors, until a new NFL season begins.
Our study considers whether such impressions may push a market to inefficiency. To search for such a “holdover bias” we examine whether lines are systematically errant when prior playoff teams face prior non-playoff teams in the following season.
For each week, we calculate the percentage of prior playoff teams which are favored when playing prior non-playoff teams. 8 We compare this to the weekly percentage of prior playoff teams that actually won these games. We test the significance of these differences with a difference of proportions z-test.
We also calculate the percentage of prior non-playoff teams which cover spreads, by week, when playing prior playoff teams. We then test if this percentage is significantly different than 52.38% via a two-sided test. This is accomplished with a one sample proportions z-test. It is common convention in sports betting literature to compare win percentages to 52.38% rather than 50.0% because, due to bookmaker commission, for a bettor to break even, based on a number of equally sized bets, he or she must win more than 52.38% of these wagers.
We also calculate line errors for all games where prior playoff teams play prior non-playoff teams. Line errors are calculated as:
Line Error = Prior Playoff Team Score + Prior Playoff Team Point Spread –
Prior Non-Playoff Team Score
8 922 of these games match prior playoff teams with prior non-playoff teams. Early Season Inefficiencies in the NFL Sports Betting Market 35
Prior Playoff Team Score and Prior Non-Playoff Team Score are the realized scores from games pitting prior playoff teams against prior non-playoff teams. Prior
Playoff Team Point Spread is the point spread relative to the team that made the playoffs in the prior year. The calculated Line Error reflects the number of points by which the spread of the game is incorrect in favor of the prior playoff team. A negative (positive) number means the playoff team was favored by too many (few) points. For example, if New England was favored by 10 points over Detroit, and the final outcome of the games was New England 24-Detroit 17, New England’s adjusted score would be 14. New England would fail to cover the spread by 3 points. The Line
Error would be -3, reflecting that New England was favored by 3 points too many.
We next perform a regression analysis which tests the efficiency of NFL lines.
We estimate the following specification as presented in Zuber et al. (1985):
Score Differencei i i
Where Score Differencei is the score of the prior playoff team less the score of the prior non-playoff team, Point Spreadi is the closing line for the game relative to the
i is the error term. If point spreads are efficient predictors of
significantly from 1. We test the coefficients from regressions for each NFL week
(aggregated over all years in the sample period) for differences from these Early Season Inefficiencies in the NFL Sports Betting Market 36
hypothesized values. As detailed by Zuber et al. (1985), statistical rejection of line efficiency, based on this specification, is difficult due to the relative lack of power of such a test.
Finally, we present specific, season by season performance of a strategy which wagers against prior playoff teams in the early weeks of a new NFL season, in order to benefit from the holdover bias of ordinary bettors. We calculate the overall, in-sample performance as well and compare the winning frequency of these wagers to the breakeven level of 52.38%. While a record better than 52.38% indicates economic profitability of the strategy, a mark heavily emphasized in the literature, we further test whether the strategy is statistically significantly better than 52.38% via a one-sample proportions z-test. We utilize a one-sided significance threshold as we hypothesize that our strategy will exceed the profitability breakeven mark. We emphasize that statistical significance of economically profitable returns would be an almost unprecedented finding for a betting market strategy, given the rather small samples that permeate the literature.
Results
Table 1 displays the performance of prior playoff teams, week by week, in the subsequent season. We consider how frequently prior playoff teams are favored, how often they win, and how often they cover the spread when playing prior non-playoff teams.
(Insert Table 1)
Early Season Inefficiencies in the NFL Sports Betting Market 37
If the betting lines create an efficient gambling market, we would expect to see no difference between the percentage of teams that are favored and the percentage of teams that win. From 2004-2011, prior playoff teams, in Week 1, are favored in 79.6% of games. However, they only win 48.1% of the time. This striking disparity means that, even with a relatively small sample size, the Week 1 difference of proportions test between percentage of prior playoff teams favored and percentage of playoff teams who win is significant at the 1% level. The analogous difference of proportions test in Week 2 is significant at the 5% level.
Prior playoff teams remain favored in more games than they actually win throughout almost all the remaining weeks of an NFL season. Table 1 notes that only in Week 17 do prior playoff teams actually win more games than they were predicted to via the gambling line, though even this result is by only 1 game (40 prior playoff teams win games while 39 were expected to, over the 2004-2011 seasons). The full sample of 922 games with one prior playoff team shows 670 teams that were favored, but only 572 of these teams actually won.9 The overall difference of proportions test is significant at the 1% level. The result is most powerful, however, in Week 1.
Given the organization of gambling markets, the more relevant question is the performance of prior playoff teams after the spread is incorporated in games the subsequent season. In Table 1, we note that in only 18 of 54 (33.3%) Week 1 games does a prior playoff team win against the spread. In 35 of 54 games the prior playoff
9 47.9% of prior playoff teams advance to the playoffs in the following year. 37.5% of teams advance to the playoffs in each season, so holdover biases, as compared to naïve assumptions that evaluate all teams equally, appear to have a basis in reality; however, the game by game projections of prior playoff team performance are too high, particularly in Week 1. Early Season Inefficiencies in the NFL Sports Betting Market 38
team loses against the spread (there is also one tie or “push” in the sample). This proportion of 66.0% of prior playoff teams failing to cover the spread (excluding the one tied game) is significantly different than 52.38% at the 1% level. This extreme result means that systematically wagering against prior playoff teams in Week 1 of an
NFL season proves to be very profitable in our sample period (further details are provided in Table 4). While we earlier observed that prior playoff teams underperform throughout the season in terms of wins versus bookmaker projections, the results against the spread are considerably different. In no week, other than Week 1, is the proportion of prior playoff teams covering significantly different than 52.38%. For the combined sample of Weeks 2-17, 49.9% of prior playoff teams cover their games against prior non-playoff teams.
We next calculate statistics regarding line errors. Table 2, below, presents the mean and median line errors for games in which exactly one of the participants is a prior playoff team. Line errors are presented which compare the actual difference in the scores of the teams to the bookmaker spreads of the games. The errors in the first column are in reference to the prior playoff team.10
(Insert Table 2)
Table 2 shows the average line error for Week 1 games in which exactly one participant is a prior playoff team is -4.69 points. This means that, on average, the prior playoff team was 4.69 points short of covering the spread (or in other words, that
10 For example, if team A is not a prior playoff team and team B is a prior playoff team, and team B is favored by 3 points, and team A wins the game by 2 points, then the line error is -2-3 = -5 points. Early Season Inefficiencies in the NFL Sports Betting Market 39
the spread is 4.69 points higher, on average, than it should be, biased toward prior playoff teams, in Week 1). This result is significantly different than 0 at the 1% level.
It is also significantly different than the average line error (0.80 points) of prior playoff teams in Weeks 2-17 at the 1% level. The results are similar when we consider the median errors. The median error is 6 points biased toward the prior playoff team, and this result is statistically different than 0, at the 5% significance level (which is the median error for games involving one prior playoff team in Weeks 2-17).
The abnormally high errors in Week 1 may be attributed to our theory that the betting market places too great an emphasis on prior season performance, which leads to inaccuracies in the spreads. Furthermore, the sharp reduction in line errors between
Weeks 1 and Weeks 2 – 17 match our expectation that this holdover bias dissipates once new information is available with which to update bettor preferences. As a brief check that the aforementioned bias in Week 1 for prior playoff teams is based in part on the Week 1 status of the game, and not entirely on the prior playoff status, we compare the line errors across other types of games in our data set. Table 2 shows that the average line error is higher in Week 1, as compared to Weeks 2-17 for all games, split into three specific categories: 1) games with no prior playoff teams; 2) games with one prior playoff team; and 3) games with 2 prior playoff teams. Higher errors for these types of games (all relative to the favorite of the game) confirm that the betting market has particular difficulty in handicapping Week 1 games.
Interestingly, the small sample of 21 observations suggests that in Week 1 games with both participants as prior playoff teams, the favorite of the game is favored Early Season Inefficiencies in the NFL Sports Betting Market 40
by too few points. We posit an explanation which coincides with our other findings.
Specifically, it appears that favorites are over-favored in Week 1 games with two prior playoff teams. It may be that the ordinary bettors of the gambling market lump all prior playoff teams too closely together in their impressions. Prior playoff status may serve as a type of dummy variable for quality, in the minds of naïve bettors, which then receives too much weight. In games with one prior playoff team, this results in overestimation of that team’s quality. In games with two prior playoff teams, this results in overestimation of an underdog’s ability, simply because it made the postseason in the prior year.11 Prior playoff status may distort the market’s perception of teams, which leads to high variability in the line errors in Week 1of the next season.
As a more formal test of gambling market statistical efficiency, we employ the regression method, first noted by Zuber et al. (1985), which regresses the final point differential of games on the spread. We utilize our sample of games involving exactly one prior playoff team and run the regression on a week-by-week basis from our sample of the 2004-2011 NFL seasons. The results are presented in Table 3.
(Insert Table 3)
The null hypothesis of an efficient gambling market can be rejected when the intercept of the regression is significantly non-zero or the slope of the regression is significantly different than 1. We are able to reject the Week 1 and Week 2 lines as
11 In most games matching two prior playoff teams, the team that advanced further in the previous year’s playoffs is the favorite. Thus, the fact that a team made the playoffs in the prior season seems to overshadow the actual gap in quality between the two teams. Early Season Inefficiencies in the NFL Sports Betting Market 41
inefficient at the 1% significance level. These are the only weeks which allow for statistical rejection at a level stronger than 10%, indicative of an inefficient gambling market in only the earliest part of an NFL season for the set of games with one prior playoff team. This statistical significance is a high threshold of inefficiency, not often realized in the literature. Zuber et al. (1985) note the regression’s statistical weakness, and Gandar et al. (1988) also fail to find statistical significance based on their trading approaches.
The more commonly cited metric of success in the literature, understandable given the motivation of gamblers, is economic profitability. A betting strategy is profitable, after factoring in bookmaker commission, if it results in winning wagers more than 52.38% of the time. Given the statistical significance noted in Table 3, we test the profitability of wagering on the opponents of prior playoff teams in Week 1 and Week 2 of an NFL season and present our results in Table 4.12
(Insert Table 4)
The results indicate substantial profitability from betting against prior playoff teams in Week 1 of the following season. In every season of our 2004-2011 sample the approach is profitable, and overall, the approach yields a profit of 25.6%.
Furthermore the statistical significance of the profitability level is strong. A one-sided test, based on the direction advocated by our strategy, rejects the null hypothesis that
12 Results of profitability for other weeks are all insignificant, as one would expect based on the against- the-spread results of Table 1. We reconsider the cases of Weeks 1 and 2 separately in Table 4 to provide further discussion. Early Season Inefficiencies in the NFL Sports Betting Market 42
the real winning percentage of this strategy is less than or equal to 52.38% at the 5% significance level. Explicitly considering the statistical significance of the economic profitability level is rare, even amongst those papers noting economic significance
(e.g., Zuber et al., 1985), and this significance is often non-existent or weak when it is considered (e.g., Gray and Gray, 1997). Finding statistical economic significance for our only hypothesis is a strong indication of an inefficient gambling market, tied to a holdover bias.
We also note that statistical significance regarding economic profitability is not redundant to the statistical significance found from the regression approach in Table 3.
Specifically, the regression approach considers whether the margins of actual game scores compare well with the margins predicted via spreads. This is an interesting statistical question, but it is a different matter to consider the question of profitability, in which each wager is an all-or-nothing proposition. In the regression framework, a large error in one game might affect parameter estimation greatly and alter statistical significance. In the economic approach, all games are of equal importance to our conclusions, regardless of the level of error between spread and actual result.
Determining statistical significance via both of these approaches adds to our confidence regarding the findings of holdover bias from one NFL season to the next.
The difference between economic significance and statistical significance from the regression framework is highlighted by our results regarding Week 2 in Table 4.
While we reject the null hypothesis of market efficiency based on the regression approach, there is no economic profitability, ex post, from betting against prior playoff Early Season Inefficiencies in the NFL Sports Betting Market 43
teams in Week 2 of an NFL season. Our findings of holdover bias appear to be only relevant in Week 1.
Out of Sample Robustness and Future Prospects
Given the unprecedented profitability over our 2004-2011 study period for the
Week 1 strategy, wagering against prior playoff teams who play prior non-playoff teams, it is natural to question whether the findings are an in-sample phenomenon, or more cynically, the result of data snooping. We utilize historical point spread data from another source, footballlocks.com, and calculate the effectiveness of the Week 1 strategy over the 1994-2003 period.
A new culture of competitive balance was developing throughout the NFL in the early 1990s. With salaries rising rapidly, players began to more frequently leave their old teams for substantial pay raises, even if this meant departing some of the most successful teams and agreeing to play for some of the least successful. 13
Furthering this shift towards increased league-wide balance, the first salary cap and floor were introduced to the NFL in 1994. These rules placed a cap on the total money used to pay player salaries by each team, as well as a minimum amount that a club could spend on player salaries. Restrictions of total team pay level were designed to place all franchises on a more equivalent footing. The NFL desired this increased parity across the league in order to garner more widespread interest and thereby
13 For a short time, the owners of NFL clubs resisted this development via the implementation of “Plan B”. In 1989, this rule gave teams the right to “protect” up to 37 players on their rosters. This protection guaranteed teams the opportunity to match any offer from another team to a current player approaching free agency and thereby retain that player’s services. This rule was found to be a violation of antitrust laws in by a U.S. Federal Court in 1992. Early Season Inefficiencies in the NFL Sports Betting Market 44
increase advertising, ticket, and merchandising revenues via broad exposure. This change manifested shortly thereafter as 12 different franchises have won the NFL championship in the 17 seasons since the salary cap’s introduction. In the 17 seasons prior, only 8 different franchises won an NFL championship.
We effectively study whether a byproduct of this dynamic, in the betting markets, has been the development of overly sticky beliefs regarding team quality, based on playoff participation, from one season to the next. From 1970-1993, with no salary cap in place, 61.2% of teams that qualified for the previous year’s postseason returned to the playoffs, so such beliefs may have been somewhat warranted. This was the case even though the NFL playoffs consisted of only 8 teams until 1978 and 10 teams from 1979-1989.14 The strength of such beliefs, however, may have been too great once the salary-cap era arrived. For our out-of-sample robustness period of 1994-
2003, the start of the salary-cap era, only 53.3% of teams that qualified for the previous year’s playoffs returned the following season. For our sample period of 2004-
2011, this figure shrinks further to 47.9%. We posit that it took some time for the salary-cap’s effects to fully emerge, with teams learning, over time, to more aggressively solicit players from opposing clubs and players more frequently seeking pay maximization.
For our out of sample robustness period of 1994-2003, the strategy of betting against prior playoff teams that face non-playoff teams in Week 1 games results in a winning rate of 54.3%, commensurate with a return of 3.6% after factoring in the
14 League membership was smaller in these seasons, but not by an amount proportional to the smaller playoff fields. Early Season Inefficiencies in the NFL Sports Betting Market 45
typical bookmaker commission. Unlike our in-sample result, this economic finding is not statistically significant; however, we note that the result still rises to the threshold of economic profitability out of sample, a standard that most studies reporting effective betting strategies (e.g. Zuber et al., 1985 and Lacey, 1990) do not reach.
Furthermore, we believe that the dynamic of the NFL betting market, going forward, sets up so that such profitability may continue. Roster turnover has increased since the implementation of the salary cap, highlighting the fallacy in believing that last year’s teams will remain intact and therefore achieve similar results in the following season.
There is little reason to believe this trend will reverse. Ordinary bettors have yet to overcome the temptation to overvalue last season’s playoff performance even though this trend has been developing for many years. We suspect they never will. Betting against prior playoff teams has actually become more profitable in the most recent
NFL seasons, our sample period. We anticipate the sticky preferences of naive bettors will continue to offer profitably opportunities for informed bettors, who are able to take contrarian positions, in the future.15
Conclusion
In our study, we consider a unique sentiment question. We hypothesize that bettors in a gambling market are overly influenced by impressions from a prior time, rather than fully adjusting their beliefs and pricing their wagers accordingly. The holdover bias, based on a classification of a team as successful because it qualified for
15 We also note that the emergence of the internet, in recent years, has made sports wagering a more acceptable and accessible activity; thus an increased number of ordinary bettors may now exist, giving more credence to the Week 1 inefficiency theory. Early Season Inefficiencies in the NFL Sports Betting Market 46
the prior season’s playoffs, allows gamblers to pay too high a price in order to cater to their biases. The structure of the betting marketplace, the overabundance of naïve participants, and the limits on entry of informed bettors allow for the existence of inefficient prices. It is particularly easy to profit from these “high price” spreads in a betting market. It is not necessary to arrange for permission to sell short or to open special accounts for margin or derivatives. Contrarian bettors need only wager against prior playoff teams. The statistical evidence confirms that bettors who take the contrarian viewpoint by simply wagering against prior playoff teams in Week 1 of an
NFL season may profit from the effectively cheap prices.
Some of the prior literature suggests that the NFL betting market does not contain any inefficiencies that can be profitably exploited by informed bettors. Those studies which have reported such inefficiencies have frequently been criticized as under-motivated, unlikely to persist out of sample, and driven by data snooping. Over the 2004-2011 NFL seasons, we find unprecedented profitability based on taking a contrarian strategy of wagering against prior playoff teams in the following season’s opening week. The line errors of these games indicate a statistically inefficient market, utilizing the low-power tests of Zuber et al. (1985). Most impressively, the profitability level of the strategy is itself found to be statistically significant, a hurdle cleared by virtually no other study of the sports betting market.
We consider only one hypothesis in our study, we develop its theoretical underpinnings through the financial and psychological literature, and we measure its performance out of sample. This out of sample testing we demonstrate that the Early Season Inefficiencies in the NFL Sports Betting Market 47
profitability of our Week 1 strategy holds, though at reduced levels. Over time, as more information about the current teams becomes available, our analysis shows a learning effect that occurs in the betting markets. As the season progresses, the line errors for games involving one prior playoff team gravitate towards the line errors observed over our entire data set.
Early Season Inefficiencies in the NFL Sports Betting Market 48
References: Avery, C., & Chevalier, J. (1999). Identifying investor sentiment from price paths: the case of football betting. Journal of Business, 72(4), 493-521.
Badarinathi, R., & Kochman, L. (1996). Football betting and the efficient market hypothesis. The American Economist, 40(2), 52-55. Barberis, N., Shleifer, A., & Vishny, R. (1998). A model of investor sentiment. Journal of Financial Economics, 49, 307-343.
Brown, G., & Cliff, M. (2005). Investor sentiment and asset valuation. Journal of Business, 78(2), 405-440.
Burkey, M. (2005). On “arbitrage” and market efficiency: An examination of NFL wagering. New York Economic Review, 36(1), 13-28.
Cooper, M., Dirmitrov, O., & Rau, P. R. (2002). A rose.com by any other name. Journal of Finance, 56(6), 2371-2388.
Dare, W. H., & Holland, A. S. (2004). Efficiency in the NFL betting market: modifying and consolidating research methods. Applied Economics , 36 (1), 9-15. Edwards, W. (1968). Conservatism in human information processing. In: Kleinmutz, B. (Ed.), Formal Representation of Human Judgment, Wiley, New York.
Fisher, K., & Statman, M. (2000). Investor sentiment and stock returns. Financial Analysts Journal, 56(2), 16-23.
Gandar, J., Zuber, R., O'Brien, T., & Russo, B. (1988). Testing Rationality in the Point Spread Betting Market. The Journal of Finance , 43 (4), 995-1008.
Golec, J., & Tamarkin, M. (1991). The degree of inefficiency in the football betting market. Journal of Financial Economics , 30, 311-323.
Gray, P. K., & Gray, S. F. (1997). Testing Market Efficiency: Evidence from NFL Sports Betting Market. The Journal of Finance , 52 (4), 1725-1737. Han, B. (2008). Investor sentiment and option prices. Review of Financial Studies, 21(1), 387- 414.
Haruvy, E., Lahav, Y., & Noussair, C. (2007). Traders’ expectations in asset markets: Experimental evidence. The American Economic Review, 97(5), 1901-1920.
Krieger, K., Fodor, A., & Stevenson, G. (2011). The sensitivity of findings of expected bookmaker profitability. Journal of Sports Economics, forthcoming.
Early Season Inefficiencies in the NFL Sports Betting Market 49
Lacey, N. (1990). An estimation of market efficiency in the NFL point spread betting market. Applied Economics 22 (1), 117-129.
Lee, M., & Smith, G. (2002). Regression to the mean and football wagers. Journal of Behavioral Decision Making, 15(4), 329-342.
Levitt, S. (2004). Why are gambling markets organized so differently from financial markets? Economic Journal 114 (495), 223-246. Millman, C. (2001). The odds: one season, three gamblers, and the death of their Las Vegas, Public Affairs (Perseus Books Group), New York, New York.
Neal, R., & Wheatley, S. (1998). Do measures of investor sentiment predict returns? Journal of Financial and Quantitative Analysis, 33, 523-547.
Pankoff, L. (1968). Market efficiency and football betting. Journal of Business, 41(2), 203-214.
Paul, R., & Weinbach, A. (2007). Does sportsbook.com set pointspreads to maximize profits? Tests of the Levitt model of sportsbook behavior. Journal of Prediction Markets, 1(3), 209-218.
Paul, R., & Weinbach, A. (2011). NFL bettor biases and price setting: further tests of the Levitt hypothesis of sportsbook behaviour. Applied Economics Letters, 18(2), 193-197.
Sapra, S. (2008). Evidence of betting market intraseason efficiency and interseason overreaction to unexpected NFL team performance 1988-2006. Journal of Sports Economics, 9(5), 488-503.
Sauer, R. D., Brajer, V., Ferris, S. P., & Marr, M. W. (1986). Holding Your Bets: Another Look at the Efficiency of the Gambling Market for National Football League Games. Journal of Political Economy , 96 (1), 206-213. Schmeling, M. (2009). Investor sentiment and stock returns: some international evidence. Journal of Empirical Finance, 16(3), 394-408.
Strumpf, K. (2003). Illegal sports bookmakers, University of North Carolina, Department of Economics.
Sturgeon, K. (1974). Guide to sports betting, Harper & Row, New York, 1974. Vergin, R. C. (1998). The NFL pointspread market revisited: anomaly or statistical aberration. Applied Economics Letters, 5(3), 175-179.
Vergin, R. C. (2001). Overreaction in the NFL point spread market. Applied Financial Economics, 11(5), 497-509.
Vergin, R. C., & Scriabin, M. (1978). Winning Strategies for Wagering on National Football League Games. Management Science , 24 (8), 809-818.
Early Season Inefficiencies in the NFL Sports Betting Market 50
Wever, S., & Aadland, D. (2012). Herd Behaviour and underdogs in the NFL. Applied Economics Letters , 19 (1), 93-97.
Woodland, B., & Woodland, L. (2000). Testing contrarian strategies in the National Football League. Journal of Sports Economics, 1(2), 187-193.
Zuber, R. A., Gandar, J. M., & Bowers, B. D. (1985). Beating the Spread: Testing the Efficiency of the Gambling Market for National Football League Games. Journal of Political Economy , 93 (4), 800-806.
Early Season Inefficiencies in the NFL Sports Betting Market 51
Table 1: Playoff Team Favorite and Win Percentages Table 1 presents statistics for each NFL week over the 2004-2011 seasons. Only games where a prior playoff team played a prior non- playoff team are included. For each week, the numbers and percentage of previous playoff teams favored to win the game and the number and percentage of prior playoffs teams who won games are presented. Differences between these proportions are shown with significance levels from a difference of proportions z-tests indicated. The results of games against the spread are also presented for each week. We exclude games where the outcome against the spread was a tie. The percentage of games where the prior playoff team won against the spread is presented with significance levels from one-sample z-tests for a winning percentage of 52.38% indicated. For conservatism, all tests are two-sided. ***, **, and * indicate significance at the 1%, 5%, and 10% levels respectively.
Favored % - Win Against Lose Against % Win Against Week N Favored % Favored Win % Win Win % Spread Spread Spread 1 54 43 79.6% 26 48.1% 31.5% *** 18 35 34.0% ** 2 64 50 78.1% 39 60.9% 17.2% ** 31 32 49.2% 3 66 52 78.8% 45 68.2% 10.6% 32 31 50.8% 4 50 38 76.0% 32 64.0% 12.0% 26 23 53.1% 5 55 42 76.4% 38 69.1% 7.3% 31 23 57.4% 6 54 43 79.6% 35 64.8% 14.8% * 27 26 50.9% 7 58 42 72.4% 34 58.6% 13.8% 28 28 50.0% 8 50 36 72.0% 31 62.0% 10.0% 26 24 52.0% 9 49 34 69.4% 25 51.0% 18.4% * 24 25 49.0% 10 55 41 74.5% 35 63.6% 10.9% 28 25 52.8% 11 59 46 78.0% 42 71.2% 6.8% 30 26 53.6% 12 60 48 80.0% 35 58.3% 21.7% *** 28 30 48.3% 13 50 37 74.0% 29 58.0% 16.0% * 22 25 46.8% 14 70 49 70.0% 45 64.3% 5.7% 37 31 54.4% 15 56 38 67.9% 32 57.1% 10.7% 27 28 49.1% 16 58 35 60.3% 35 60.3% 0.0% 32 25 56.1% 17 68 39 57.4% 40 58.8% -1.5% 27 38 41.5% 2-17 922 670 72.7% 572 62.0% 10.6% *** 456 440 50.9% *** All 976 713 73.1% 598 61.3% 11.8% 474 475 49.9% Early Season Inefficiencies in the NFL Sports Betting Market 52
Table 2: Line Errors Table 2 presents mean and median line errors after separating games according to the number of teams in the game that made the playoffs in the prior year. For the one-playoff team sample, line errors are calculated relative to the prior playoff team as prior playoff team score, less prior non- playoff team score, adjusted for the spread of the game relative to the prior playoff team. A negative line error therefore indicates that prior playoff teams were favored by too many points (or were underdogs by too few points) and fell short of covering the spread by this amount. A positive line error indicates that prior playoff teams were favored by too few points (or were underdogs by too many points) and covered the spread by this amount. Line errors for the no- playoff and two-playoff team samples are calculated relative to the favorite team as the favorite team score, less the underdog team score, less the number of points the favorite team is favored by. A negative line error therefore indicates the favorite team was favored by too many points and fell short of covering the spread by this amount. A positive line error indicated that favorite team was favored by too few points and covered the spread by this amount. Significance levels from t-tests for means and signed-rank tests for medians are indicated. Results are presented separately for Week 1 and Weeks 2-17. Differences between Week 1 and Weeks 2-17 line errors are presented with significance levels from two sample t-tests for mean differences and Wilcoxon rank-sum tests for median differences. The sample period is the 2004-2011 NFL seasons. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.
Panel A: Mean Line Errors
One Playoff Team No Playoff Teams Two Playoff Teams N Mean N Mean N Mean *** ErrorWeek 1 54 -4.69 52 -0.18 21 4.29
ErrorWeeks 2-17 922 0.80 736 0.26 259 -1.17 *** * ErrorWeek 1 - ErrorWeeks 2-17 -5.49 -0.44 5.46
Panel B: Median Line Errors
One Playoff Team No Playoff Teams Two Playoff Teams N Median N Median N Median ** ErrorWeek 1 54 -6.00 52 -0.75 21 4.00
ErrorWeeks 2-17 922 0.00 736 0.00 259 -1.00 *** * ErrorWeek 1 - ErrorWeeks 2-17 -6.00 -0.75 5.00 Early Season Inefficiencies in the NFL Sports Betting Market 53
Table 3: Regression Analysis Table 3 presents regression results for each NFL week over the 2004-2011 seasons. The estimated model regresses the realized point differential of games involving one prior playoff team, relative to the prior playoff team, on the spread of the game. Only games where a prior playoff team played a prior non-playoff team are included. Intercepts and coefficients of spreads are presented. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.
Week N Intercept Line 1 54 -5.999 *** 1.358 2 64 -4.659 ** 2.005 *** 3 66 3.202 0.769 4 50 1.206 0.862 5 55 3.531 0.828 6 54 2.052 0.782 7 58 1.424 0.749 8 50 1.640 0.643 9 49 -1.710 0.890 10 55 1.633 0.751 11 59 2.222 1.221 12 60 -3.895 * 1.709 * 13 50 -0.640 1.207 14 70 0.815 1.347 15 56 -0.696 1.139 16 58 3.494 * 0.733 17 68 -0.990 1.205 2-17 922 0.605 1.052 All 976 0.259 1.064
Early Season Inefficiencies in the NFL Sports Betting Market 54
Table 4: Profitability of Betting Against Prior Playoff Teams Table 4 presents the results of prior playoff teams when playing against prior non- playoff teams for our sample period of the 2004-2011 NFL seasons. The sample consists of Week 1 and Week 2 games and is divided according to season. Against- the-spread results are presented for wagering on the prior non-playoff teams. The percentage of games where the prior playoff teams loses against the spread is calculated excluding pushes. Due to bookmaker commission, on a bet of $110, a win pays $210, a push pays $110 and a loss pays nothing; thus, to be profitable, wagers must win 52.38% of the time. Statistical significance of betting against prior playoff teams in weeks 1 and 2 is measured by comparing winning percentages to 52.38%. Statistical significance is tested for total Week 1 and Week 2 results only, given the small sample sizes for individual years and the nature of our strategy. * denotes statistical significance at the 5% level We utilize one-sided significance levels due to our posited findings. % Return is calculated as total return on equal bets on each game over the total invested and incorporates the bookmaker commission. Panel A: Week 1
Lose Against Win Against % Lose Against Season N Spread Push Spread Spread % Return 2004 6 4 0 2 66.7% 27.3% 2005 10 7 0 3 70.0% 33.6% 2006 10 6 0 4 60.0% 14.5% 2007 4 3 0 1 75.0% 43.2% 2008 8 5 0 3 62.5% 19.3% 2009 6 3 1 2 60.0% 12.1% 2010 4 3 0 1 75.0% 43.2% 2011 6 4 0 2 66.7% 27.3% Total 54 35 1 18 66.0% ** 25.6%
Panel B: Week 2
Lose Against Win Against % Lose Against Season N Spread Push Spread Spread % Return 2004 8 3 0 5 37.5% -28.4% 2005 8 4 0 4 50.0% -4.5% 2006 10 4 0 6 40.0% -23.6% 2007 6 5 0 1 83.3% 59.1% 2008 12 4 1 7 36.4% -28.0% 2009 6 3 0 3 50.0% -4.5% 2010 8 5 0 3 62.5% 19.3% 2011 6 4 0 2 66.7% 27.3% Total 64 32 1 31 50.8% -3.0% Early Season Inefficiencies in the NFL Sports Betting Market 55
Early Season NFL Over/Under Bias
Michael DiFilippo Justin Davis Ohio University Ohio University 409 Copeland Hall Department of Management Athens, OH 45701 310 Copeland Hall phone: 740-593-2055 Athens, OH 45701 fax: 740-593-9342 phone: 740-593-2391 [email protected] fax: 740-593-9342 [email protected]
Andy Fodor* Kevin Krieger Ohio University University of West Florida Finance Department Department of Accounting and Finance 234 Copeland Hall Bldg 76, Rm 226 Athens, OH 45701 Pensacola, FL 32514 phone: 740-593-0514 Phone: 850-474-2720 fax: 740-593-9342 Fax: 850-474-2717 [email protected] [email protected]
Early Season Inefficiencies in the NFL Sports Betting Market 56
Abstract
Popular wisdom regarding athletics is that offenses are at a relative disadvantage in the early portion of seasons. We present evidence that this anecdotal belief holds true over the 2000-2010
National Football League (NFL) seasons. This is reflected in lower offensive yardage, fewer first downs, and fewer points scored. While total points scored are significantly lower in Week 1 of NFL seasons, bookmakers fail to reduce the total lines posted on these games. We find a strategy betting under total lines of all Week 1 games over the 2000-2010 NFL season yields a statistically significant profit of 13.6% per game.
Keywords: Sports Wagering, Efficient Markets, NFL, Betting Bias Early Season Inefficiencies in the NFL Sports Betting Market 57
Introduction
Sports journalists and analysts often assert that defenses have advantages over offenses in early season play. The most common cliché professing this belief is that “defenses are ahead of the offenses early.”16 Underlying this theory is the belief that offenses in athletics require more practice repetition in order to hone execution to optimal levels while defenses rely more on instinct and athleticism in order to stifle offenses. We consider whether the empirical evidence supports these assertions in the National Football League (NFL).
We find evidence that offenses are, indeed, relatively slow to produce in the first week of an NFL season. Offensive yardage and number of first downs registered are significantly lower in the opening week of NFL seasons than at any other time. This results in fewer points scored in
Week 1 of NFL seasons than in any other week.
Offenses do perform poorly early in seasons, but does the public fully realize the degree of this disparity? To answer this question we consider whether poor offensive performance is fully accounted for in NFL betting markets. According to efficient market theory, gambling lines established for NFL games should incorporate all available information. This includes the tendency of Week 1 NFL games to have significantly lower total scores than games in later weeks. If total lines do not adjust to reflect differences in points scored in Week 1 relative to later weeks this is evidence of inefficiency in NFL betting markets.
16 For example: http://www.howardbison.com/sports/fball/200910/releases/Defense_Stands_Out_In_First_Scrimmage http://www.thepilot.com/news/2010/aug/14/defense-ahead-offense-scrimmages/ http://www.mysoutex.com/view/full_story_landing/282084/article-Trojan-defense-ahead-of--O--in-early-practices- Scrimmages-set-for-Saturday-morning-for-all-Bee-County-teams http://www.jimfeist.com/editorials-by-sport.html?leagueid=NCAAF&editorialid=1173 Early Season Inefficiencies in the NFL Sports Betting Market 58
The rationality/efficiency of sports betting markets has been a subject of debate in the literature for many years. Numerous authors have documented profitable betting strategies or rules of thumb, suggesting inefficiencies in sports betting markets. Much of this work involves the NFL, which has become the most widely viewed and highly wagered upon segment of
American sports.
In early studies, Sturgeon (1974) and Vergin and Scriabin (1978) describe how contrarian positions, for example, wagering against the best NFL teams, could in fact yield profits. More general betting approaches, based on a number of team statistics, are advocated by Zuber et al.
(1985) and Gandar et al. (1988), who use regressions to determine the importance of the various measures and utilize the coefficients of the measures to forecast margins of victory. They then advocate wagering against the spread when bookmaker lines vary dramatically from these predictions. More recent research has again advocated the use of contrarian positions in order to earn profits. Golec and Tamarkin (1991) describe how betting on underdogs can be profitable, and in new research, Paul and Weinbach (2011) and Wever and Aadland (2012) find that wagering specifically on large underdogs is a profitable position.
While many authors identify wagering strategies that appear profitable, the viability of these strategies has come under scrutiny from critics. When authors have explored numerous strategies simultaneously, with little theoretical development for their potential existence, the critiques have been particularly harsh. Burkey (2005) notes that authors who consider enough trading rules will definitely uncover signs of profitability after the fact. He suggests that a well- developed theory, ex ante, may serve to appreciably improve the credibility of findings. As targets, he notes that Woodland and Woodland (2000) reject 7 null hypotheses which individually test that the NFL betting market is efficient; however, they utilize statistical Early Season Inefficiencies in the NFL Sports Betting Market 59
significance at the 10% level to do so and study 48 different hypotheses. Additionally,
Badarinathi and Kochman (1996) reject 7 of 116 null hypotheses, each testing the betting market’s efficiency.
In this paper we consider one new potential inefficiency of the NFL betting market, which is suggested by popular belief and further advanced by our initial analysis. Specifically, in the early weeks of an NFL season, we hypothesize that totals of games will systematically be set too high and betting under total lines in Week 1 will result in a profitable betting strategy.17 We find that betting under the totals of Week 1 games results in returns of 13.6% per game on average over the 2000-2010 NFL seasons. As well as being economically profitable, these returns are statistically significant, a threshold reached by very few studies which consider the efficiency of gambling markets.18
Data, Methodology, and Results
The sample for our study consists of all NFL regular season games from the 2000-2010 seasons. Total line data is from Sports Insight. The total line is a betting line set by bookmakers allowing bettors to wager that the total score for the game will be above or below this total line.
A bet of $110 will pay $210 if correct and nothing if incorrect. Offensive first down, yardage, and points scored data is from Sunshine Forecast.
17 “Totals” reflect the total number of points that oddsmakers project to be scored in a contest. Bettors may wager that the number of points scored between the two teams will go over (under) this number, oftentimes because bettors expect a game to have more (less) successful offensive play than the total would suggest. For this reason a total is often referred to colloquially as an “over/under”. 18 For example, the high profile study of Gandar et al. (1988) notes: “None of the mechanical rules can be considered profitable based on these (statistical) criteria…while several achieve higher winning-bet percentages than needed to break even, none of these have Z-values sufficiently large to reject the null hypothesis of randomness and unprofitability at conventional levels of significance.” Early Season Inefficiencies in the NFL Sports Betting Market 60
We first calculate means and medians for offensive points, offensive yardage and offensive first downs for each team in a game after dividing the sample by week. We next present analysis comparing total lines and total points scored after the dividing the sample according to week. Last we examine the success and profitability of a strategy that bets under the total line on all games. This is examined by week as well as for each individual season from
2000-2010.
In Table 1 Panel A (B), mean offensive points, yardage, and first downs are presented for the sample of all NFL regular season games over the 2000-2010 season, each individual week, and weeks 2-17. The results show that NFL offenses perform relatively poorly early in seasons.
Specifically, points, yardage and first downs are lower in the first week of seasons than in later weeks. The average number of points scored per team is 19.63 in Week 1 games relative to
21.33 points per team on average for games later in the season. This difference is significant at the 1% level. Differences between average Week 1 points and average points in all other individual weeks are also calculated. Average points are lower in Week 1 than for each other individual week. 13 of the 16 differences between the average Week 1 score and later week scores are significance at the 10% level or better.
Lower points scored in Week 1 can be explained by relative low offensive yardage and fewer first downs on average in Week 1. Average offensive yardage per team in Week 1 is 323.4 compared to an average of 337.2 yards in other weeks. This difference is significant at the 1% level. Average yardage is lower in each of the other 16 weeks compared to Week 1 and 13 of the
16 differences are significant at the 10% level or better. For first downs, the average for Week 1 is 17.63 per team compared to 18.60 in other weeks. The difference is again significant at the
1% level. First downs are also lower in Week 1 than in each other individual week with 14 of 16 Early Season Inefficiencies in the NFL Sports Betting Market 61
differences significant that 5% level or better. For points, yardage and first downs, median results are nearly identical to mean results, showing the finding presented in Panel A are not driven by a small number of games were a team (teams) exhibit especially poor offensive performance. This provides evidence that defense are ahead in their preparation for the beginning of seasons compared to offenses. We next test for inefficiency in total line setting across weeks.
In Table 2 Panel A (B) mean (median) total lines and total points are presented for all games in the 2000-2010 NFL regular seasons. Results are presented for the sample of all NFL regular season games, each individual week, and weeks 2-17. In Week 1, the mean total score is less than the mean total line by 1.99 points. The median total score is less than the median total line by 3.5 points. These differences are significant at the 10% and 5% levels respectively.
Further, mean and median difference between Week 1 total lines and totals lines in weeks 2-17 are insignificant while total score differences between Week 1 and weeks 2-17 are significant at the 1% level for both means and medians. These results suggest inefficiency in Week 1 total line setting.
Given the findings presented in Tables 1 and 2, we examine if a strategy of betting on games to go under total lines is profitable in Week 1 is profitable. If total lines do not take into account the tendency of offenses to perform relatively poorly in Week 1, profit should be possible from such a strategy. Table 3 presents the number of games where total points are over and under the total lines for all games in the 2000-2010 NFL regular seasons excluding those games where the result was a push. Results are presented for the sample of all NFL regular season games, each individual week, and weeks 2-17. In Week 1, total points are lower than total lines in 102 of 171 games in the sample. This is much higher than the 52.38% of successful Early Season Inefficiencies in the NFL Sports Betting Market 62
bets required for a profitable betting strategy. Further, this percentage is significantly different from 52.38% at the 10% level. Not surprisingly, the percentage of games where the total score is less than the total line is almost precisely 50% (49.7%). The finding of significant profitability for a strategy that bets total scores will be less than total lines suggests bookmakers do not adjust total lines in Week 1 to account for relatively poor offensive performance.
In Table 4 we directly test for the presence and consistency of this inefficiency by presenting the results of betting total scores in each Week 1 game will fall below the total line.
Results are presented for all Week 1 games in the 2000-2010 NFL regular seasons and for each individual year from 2000-2010. Over the full sample, the strategy yields an average return of
13.6% per year and is profitable in 9 of 11 seasons. Overall, the results suggest NFL offenses perform poorly in Week 1 relative to the NFL season. However, bookmakers do not adjust total lines to reflect this deficiency, leading to an inefficiency that can be exploited by bettors.
Conclusion
We document a rare, statistically significant, inefficiency in NFL betting markets.
Following the notion that offenses are behind defenses in preparation early in NFL seasons, we examine Week 1 offensive performance relative to performance in other weeks. We find offenses do perform relatively poorly in Week 1. This is reflected in lower offensive yardage, fewer first downs, and fewer points scored.
While these trends exist over the length of our sample period (2000-2010 NFL regular seasons), bookmakers have failed to adjust lines. Points scored are significantly lower by 3.4 points in Week1 than in later weeks while total lines show small (0.4 points) insignificant differences. Given this, it is not surprising that we find 59.6% of games in Week 1 have total score less than total lines. This percentage is significantly higher than the 52.38% win Early Season Inefficiencies in the NFL Sports Betting Market 63
percentage needed for a profitable betting strategy. In later weeks 49.7% of games have total scores less than total lines.
The strategy betting under total lines in Week 1 yields an average profit of 17.6% per game over the 2000-2010 NFL seasons. Further, this strategy is profitable in 9 of the 11 seasons examined. If bookmakers fail to adjust Week 1 lines to reflect the relatively poor performance of
NFL offenses in Week 1, a continued opportunity may exist to profit from this inefficiency in
NFL betting markets.
Early Season Inefficiencies in the NFL Sports Betting Market 64
References:
Avery, C., & Chevalier, J. (1999). Identifying investor sentiment from price paths: the case of football betting. Journal of Business, 72(4), 493-521.
Badarinathi, R., & Kochman, L. (1996). Football betting and the efficient market hypothesis. The American Economist, 40(2), 52-55.
Burkey, M. (2005). On “arbitrage” and market efficiency: An examination of NFL wagering. New York Economic Review, 36(1), 13-28.
Gandar, J., Zuber, R., O'Brien, T., & Russo, B. (1988). Testing Rationality in the Point Spread Betting Market. The Journal of Finance , 43 (4), 995-1008.
Golec, J., & Tamarkin, M. (1991). The degree of inefficiency in the football betting market. Journal of Financial Economics , 30, 311-323.
Krieger, K., Fodor, A., & Stevenson, G. (2011). The sensitivity of findings of expected bookmaker profitability. Journal of Sports Economics, forthcoming.
Levitt, S. (2004). Why are gambling markets organized so differently from financial markets? Economic Journal 114 (495), 223-246. Strumpf, K. (2003). Illegal sports bookmakers, University of North Carolina, Department of Economics.
Sturgeon, K. (1974). Guide to sports betting, Harper & Row, New York, 1974.
Vergin, R. C., & Scriabin, M. (1978). Winning Strategies for Wagering on National Football League Games. Management Science , 24 (8), 809-818.
Wever, S., & Aadland, D. (2012). Herd Behaviour and underdogs in the NFL. Applied Economics Letters , 19 (1), 93-97.
Woodland, B., & Woodland, L. (2000). Testing contrarian strategies in the National Football League. Journal of Sports Economics, 1(2), 187-193.
Zuber, R. A., Gandar, J. M., & Bowers, B. D. (1985). Beating the Spread: Testing the Efficiency of the Gambling Market for National Football League Games. Journal of Political Economy , 93 (4), 800-806.
Early Season Inefficiencies in the NFL Sports Betting Market 65
Table 1: Offensive Statistics Table 1 presents mean and median offensive points, offensive yardage and number of first downs, by week, for weeks 2-17, and for all games in Panels A and B, respectively. Means and medians are calculated using statistics for each team in a game individually. Results are presented over the 2000-2010 NFL seasons. Mean and median differences between Week 1 and each individual week as well as Week 1 and weeks 2-17 are also presented. Significance levels from two-sided t-tests for mean differences and two-sided Wilcoxon rank-sum tests for median differences are indicated. ***, **, and * indicate significance at the 1%, 5%, and 10% levels respectively.
Panel A: Means Offensive Difference Offensive Difference Offensive Difference Week Points from Week 1 Yards from Week 1 First Downs from Week 1 1 19.63 NA 323.4 NA 17.63 NA 2 20.75 1.12 337.7 14.2 *** 18.44 0.80 ** 3 21.08 1.45 * 338.2 14.8 *** 18.77 1.14 *** 4 21.35 1.72 ** 334.8 11.4 * 18.53 0.89 *** 5 22.09 2.47 *** 338.2 14.8 *** 18.86 1.23 *** 6 20.92 1.30 * 341.2 17.7 *** 18.57 0.94 *** 7 22.37 2.75 *** 340.4 17.0 *** 18.85 1.22 *** 8 21.48 1.85 *** 337.1 13.6 * 18.63 0.99 *** 9 21.22 1.60 ** 333.5 10.1 18.86 1.23 *** 10 22.00 2.37 *** 342.7 19.3 *** 18.75 1.12 *** 11 20.39 0.77 338.1 14.7 *** 18.53 0.90 *** 12 21.88 2.25 *** 342.8 19.4 *** 19.02 1.39 *** 13 21.42 1.79 ** 337.0 13.6 ** 18.53 0.90 *** 14 20.74 1.11 333.2 9.7 18.21 0.58 15 21.46 1.84 *** 335.5 12.1 ** 18.50 0.87 *** 16 21.04 1.41 * 335.6 12.1 * 18.58 0.94 *** 17 21.27 1.65 ** 330.1 6.6 18.08 0.45 2-17 21.33 1.70 *** 337.2 13.8 *** 18.60 0.97 *** All 21.22 NA 336.3 NA 18.54 NA
Early Season Inefficiencies in the NFL Sports Betting Market 66
Table 1 cont. Panel B: Medians Offensive Difference Offensive Difference Offensive Difference Week Points from Week 1 Yards from Week 1 First Downs from Week 1 1 19.5 NA 316.5 NA 18.00 NA 2 20 0.50 336.0 19.5 ** 18.00 0.00 3 21 1.50 * 336.0 19.5 ** 19.00 1.00 *** 4 20 0.50 * 340.0 23.5 ** 19.00 1.00 ** 5 21 1.50 *** 339.5 23.0 ** 19.00 1.00 *** 6 21 1.50 * 337.0 20.5 *** 19.00 1.00 ** 7 21 1.50 *** 341.0 24.5 *** 19.00 1.00 *** 8 21 1.50 ** 347.5 31.0 ** 19.00 1.00 ** 9 21 1.50 ** 333.0 16.5 19.00 1.00 *** 10 21.5 2.00 *** 342.0 25.5 *** 19.00 1.00 *** 11 20 0.50 341.5 25.0 ** 19.00 1.00 ** 12 21 1.50 *** 341.0 24.5 *** 19.00 1.00 *** 13 21 1.50 ** 338.0 21.5 ** 18.50 0.50 *** 14 20 0.50 331.5 15.0 * 18.00 0.00 15 21 1.50 ** 336.0 19.5 ** 19.00 1.00 ** 16 21 1.50 ** 338.5 22.0 ** 19.00 1.00 ** 17 21 1.50 ** 330.0 13.5 18.00 0.00 2-17 21 1.50 *** 338.0 21.5 *** 19.00 1.00 *** All 21 NA 337.0 NA 19.00 NA
Early Season Inefficiencies in the NFL Sports Betting Market 67
Table 2: Total Line Errors Table 2 presents mean and median total scores and total lines by week, for weeks 2-17, and for all games. Results are presented over the 2000-2010 NFL seasons. Mean and median differences between total scores and total lines are presented in Panels A and B respectively with significance levels from two-sided t-tests for mean differences and Wilcoxon rank-sum tests for median differences. ***, **, and * indicate significance at the 1%, 5%, and 10% levels respectively.
Panel A: Means Panel B: Medians Week N Total Line Total Score Difference Week N Total Line Total Score Difference 1 174 41.2 39.3 1.99 * 1 174 41.0 37.5 3.50 ** 2 172 41.0 41.5 -0.51 2 172 41.0 40.0 1.00 3 162 41.3 42.2 -0.82 3 162 41.5 41.0 0.50 4 154 41.2 42.7 -1.52 4 154 41.5 42.0 -0.50 5 154 41.2 44.2 -3.03 *** 5 154 41.3 45.0 -3.75 *** 6 151 41.8 41.8 -0.05 6 151 42.5 41.0 1.50 7 152 41.7 44.7 -3.03 *** 7 152 41.8 44.0 -2.25 8 151 41.7 43.0 -1.21 8 151 41.5 42.0 -0.50 9 153 41.8 42.4 -0.61 9 153 41.0 43.0 -2.00 10 159 41.8 44.0 -2.21 * 10 159 41.5 44.0 -2.50 11 174 42.1 40.8 1.33 11 174 41.8 41.0 0.75 12 174 42.0 43.8 -1.80 12 174 42.0 42.5 -0.50 13 174 41.9 42.8 -0.98 13 174 42.0 43.0 -1.00 14 174 41.7 41.5 0.19 14 174 41.3 41.0 0.25 15 175 41.4 42.9 -1.55 15 175 41.0 42.0 -1.00 16 173 41.7 42.1 -0.38 16 173 41.0 42.0 -1.00 17 174 40.9 42.5 -1.58 17 174 40.8 41.0 -0.25 2-17 2,626 41.6 42.7 -1.08 *** 2-17 2,626 41.5 42.0 -0.50 * All 2,800 41.5 42.4 -0.89 *** All 2,800 41.5 41.5 0.00 Early Season Inefficiencies in the NFL Sports Betting Market 68
Table 3: Over/Under Results Table 3 presents the number of games with total scores over and under the total line, by week, for weeks 2-17, and for all games. Results are presented over the 2000-2010 NFL seasons. The percentage of games where the total score was below the total line is also presented by week. Significance levels from two-sided t-tests for differences from 52.38% (when the percentage of games going under the total line is greater than 52.38%) or 47.62% (when the percentage of games going under the total line is less than 47.62%) are indicated by asterisk. * indicates significance at the 10% level.
Week N Under Over % Under 1 171 102 69 59.6% * 2 167 88 79 52.7% 3 160 76 84 47.5% 4 149 73 76 49.0% 5 152 61 91 40.1% * 6 150 77 73 51.3% 7 148 66 82 44.6% 8 148 71 77 48.0% 9 148 75 73 50.7% 10 156 71 85 45.5% 11 169 97 72 57.4% 12 171 93 78 54.4% 13 174 86 88 49.4% 14 169 94 75 55.6% 15 169 79 90 46.7% 16 170 86 84 50.6% 17 174 86 88 49.4% 2-17 2,574 1,279 1,295 49.7% All 2,745 1,381 1,364 50.3%
Early Season Inefficiencies in the NFL Sports Betting Market 69
Table 4: Profitability of Betting Under the Total Line Table 4 presents the results of over/under bets in Week 1. Results are presented for each NFL season from 2000 through 2010. % Return is calculated as total return based on equal bets under the total of each game relative to the total amount bet. On a bet of $110, a win pays $210, a push pays $110 and a loss pays nothing.
Season N Under Push Over % Return 2000 15 10 0 5 27.3% 2001 15 10 0 5 27.3% 2002 16 4 0 12 -52.3% 2003 16 9 0 7 7.4% 2004 16 9 0 7 7.4% 2005 16 9 2 5 19.9% 2006 16 12 0 4 43.2% 2007 16 11 0 5 31.3% 2008 16 9 0 7 7.4% 2009 16 8 0 8 -4.5% 2010 16 11 1 4 37.5% Total 174 102 3 69 13.6%
Early Season Inefficiencies in the NFL Sports Betting Market 70
Expanded Literature Review
The efficiency of the NFL sports betting market has been a topic of investigation in academics for over forty years. Given that this market has grown into a multi-billion dollar arena for gambling, the primary stream of research has been aimed at identifying inefficiencies that are exploitable for financial gain. To this point, the findings from these efforts have been mixed. Many authors have identified trends in the betting markets that, if exploited over certain periods, would have led to significant profits. However, other authors have often found such results fail to persist out of sample. Some of the findings of systematic profitability generate from strategies lacking a strong underlying theory, thus exacerbating data mining concerns.
The articles written by my co-authors and me identifies profitable betting strategies that take advantage of inefficiencies in the NFL sports betting market during the first week of the NFL season. The biases we discovered add to the literature reviewed below by establishing two, simple, statistically significant strategies for bettors that exploit inefficiencies in the NFL market in the first week of the season.
As discussed in the following literature review, numerous past works have strived to find statistically significant and highly profitable betting strategies in the
NFL market. However, very few, if any, of the papers have been able to find a strategy that satisfies all of the requirements of being statistically significant, reoccurring and profitable. Our article stands as a significant addition to the NFL sports betting literature through its unique, profitable and intuitive strategies.
Early Season Inefficiencies in the NFL Sports Betting Market 71
Pankoff (1968) was the initial study of efficiency in the NFL betting market and studied point spreads from all 856 NFL games for the ten-year period from 1956 through 1965. Pankoff sought to answer the questions of “1) does the market have patterns; 2) if so, are they large enough to be profitable to a bettor; and 3) even in the absence of patterns, is it possible for superior analysts to beat the market consistently?” Using linear regression analysis, Pankoff was able to conclude,
“straightforward, systematic market error patterns [in the NFL market] are not large enough to be profitable to bettors.” Pankoff’s findings showed that the theory of efficient markets applies in the NFL gambling market, concluding that no significant efficiencies exist.
Vergin and Scriabin (1978) sought to refute Pankoff (1968) by not only attempting to find inefficiencies in the NFL market, but also by developing profitable strategies to exploit the inefficiencies. Vergin and Scriabin sought out profitable strategies by studying a number of betting rules of thumb. Through analyzing all of the
NFL regular season games over the 1969 through 1974 seasons, Vergin and Scriabin concluded that by betting on the underdog team when the spread is greater than 5 points, a bettor would have generated an average profit of 5% per game over the 1969 through 1974 seasons.
Vergin and Scriabin also analyzed the strategies outlined in Sturgeon (1974)’s book entitled “Sports Betting Guide.” These strategies included: 1) betting on teams that consistently beat the point spread over the previous season, or conversely betting against those that had performed poorly against the spread in previous season and 2)
Early Season Inefficiencies in the NFL Sports Betting Market 72
betting on the NFL team which won its game by the biggest margin the previous weekend. Vergin and Scriabin concluded that neither strategy proved to be significantly profitable for bettors. Vergin and Scriabin’s main contribution to the literature was finding their “underdog strategy” profitable which disproved Pankoff
(1968) and showed that inefficiencies exist in NFL betting markets.
Zuber et al. (1985) sought to further investigate the findings of Vergin and
Scriabin (1978) by taking a different approach to predicting NFL game outcomes.
Zuber et al. focused on series of individual team performance and characteristic measures to predict the outcome of games. Some of the metrics that Zuber et al. focused on included fumbles, interceptions, number of penalties, proportion of passing plays attempted relative to total offensive plays and number of rookies. Through developing an equation that accounts for all of these variables, Zuber et al. was able to generate a profitable method of predicting the outcome of games.
After using their prediction equation to place simulated bets during the 1983
NFL regular season, they were able to generate a winning percentage of 59 percent, which is significant at the 5 percent level. It can be concluded that Zuber et al. conditionally affirmed the findings of Vergin and Scriabin (1978) that profitable betting strategies can be developed to exploit inefficiencies in the NFL sports betting market. The findings of Zuber et al, must be taken with caution as their study was conducted over a small sample size and therefore “simulations over additional seasons are required before confidence in these results can be assured.”
Early Season Inefficiencies in the NFL Sports Betting Market 73
Sauer et al. (1986) sought to refute the findings of Zuber et al. (1985) by describing how wagering based on the findings of Zuber et al. leads to substantial losses in out of sample testing. Sauer et al. argues that the tests performed by Zuber et al. are misleading and that the Zuber et al. adds no new information to that already included in the Las Vegas lines. These weaknesses as discussed by Sauer et al. are further affirmed when Sauer et al. reproduces the findings of Zuber et al. in out of sample testing and finds the strategies lead to losses. Sauer et al. ultimately concludes that Zuber et al. “fails to provide sufficient evidence to support the argument that speculative inefficiencies exist in the betting market for NFL games.” Sauer et al. further enforces the difficulty in developing consistent profitable betting strategies for the NFL sports betting market.
Expanding upon the findings of Vergin and Scriabin (1978), Gandar et al.
(1988) introduced new rules of thumb to study. In addition to the rules tested by
Vergin and Scriabin, Gandar et al. tests 3 additional behavioral rules that include 1) bet on the team that becomes less favored over the course of the week’s betting; 2) bet against the public for games in weeks following “winning” weeks for the public; and
3) bet the underdog against a favored team that, as a favorite in the previous week, covered the spread by at least 10 points. Gandar et al. concludes that none of Vergin and Scriabin’s rules can be considered profitable on a statistically significant basis.
However, they do conclude that their 3 additional rules are profitable, further refuting the findings of Pankoff (1968) by indicating inefficiencies do exist in the NFL betting market.
Early Season Inefficiencies in the NFL Sports Betting Market 74
Lacey (1990) examined the NFL betting market over the 1984-1986 seasons.
Utilizing 13 technical rules for betting, Lacey found several that had greater than 50% winning percentages and were profitable after transaction costs. Vergin (1998) finds that profitable trading rules developed by Lacey, based on the 1984-1986 NFL seasons, did not hold for the subsequent 1987-1995 period. Vergin concludes that
Lacey’s findings were the result of statistical aberration and do not reveal a violation of the weak form of the Efficient Markets Hypothesis.
Golec and Tamarkin (1991) continue the work of previous literature by searching for statistically significant, profitable betting strategies. In their analysis, data from the 1973-1987 seasons is used to test betting strategies including betting solely for home teams, betting for underdogs and betting on home underdogs. Golec and Tamarkin conclude that betting on NFL home underdogs is a statically significant winning strategy with a winning percentage greater than the required profitable percentage of 52.38%. Golec and Tamarkin affirm the findings of the previous literature that inefficiencies do exist in the NFL market. However, they acknowledge that their strategy of betting for home team underdogs offers only a marginal profit and only applies for Las Vegas-type cost conditions.
Badarinathi and Kochman (1996) test three betting strategies that were touted as profitable by Tryfos et al. (1984). Each of these three strategies called for betting on the underdog when the point spread was greater than five points. The variation in the three points spreads is attributed to different advantages available to the bettor through the use of syndicates. The three betting strategies specifically state the size of the
Early Season Inefficiencies in the NFL Sports Betting Market 75
advantage as part of the strategy (1, 1.5 or 2.0 points). Tryfos et al. sought to investigate how Vergin and Scriabin (1978) measured profitability and whether their findings were sample-specific. Tryfos et al. found that the testing methods used by
Vergin and Scriabin were inaccurate as they were not specifically profit based.
Therefore, Tryfos et al. retested the 23 strategies outlined in Vergin and Scriabin and confirmed the previous findings. However, when Tryfos et al. applied the strategies of
Vergin and Scriabin over the 1975-1981 NFL seasons, only three of the 23 rules were profitable.
Badarinathi and Kochman (1996) took these three rules and applied them to the 2,272 regular and post-season games that were played during the 1984-1993 NFL seasons. They found the three strategies showed mixed profitability over their data set.
Badarinathi and Kochman conclude that regular profits are possible by betting on NFL underdogs who receive 5 ½ points or more from bookmakers and no fewer than two points from syndicates. Badarinathi and Kochman affirm the findings in NFL betting literature that inefficiencies do in fact exist in the NFL betting market.
Gray and Gray (1997) find in-sample profitability based on trading rules but acknowledge, in line with the findings of Sauer et al. (1986), that results are considerably mixed out of sample. Gray and Gray replace the standard ordinary least squares regression methodology with a probit model that enables the test to incorporate sophisticated betting strategies where bets are placed only when there is a relatively high probability of success. Gray and Gray find that the strategy of betting on home underdogs has limited success with average returns over 4 percent in excess
Early Season Inefficiencies in the NFL Sports Betting Market 76
of commissions in only 3 out of 11 seasons that were tested. Gray and Gray further conclude that other strategies tend to lose profitability over time as the apparent inefficiencies documented dissipate over time.
Vergin (1998) examines the trading rules developed by Lacey (1990) over the
1987-1995 NFL season. Lacey developed and tested 13 technical rules for betting and found several of them to be profitable after transaction costs. The finding of profitable strategies by Lacey indicated that the Efficient Markets Hypothesis does not apply to
NFL sports betting markets. However, Vergin (1998) refutes the findings of Lacey by finding none of the strategies were profitable over the 1987-1995 seasons. Vergin, similar to Sauer (1986), demonstrates how difficult it for bettors to find betting strategies that are profitable season-over-season. Furthermore, Vergin adds to the literature by contending that the Efficient Market Hypothesis remains valid for NFL betting markets.
Woodland and Woodland (2000) contributes to the literature by extending the analysis of contrarian strategies from the stock market to the NFL sports betting market. Woodland and Woodland contend that bettors, similar to investors, can overreact to any game related information such as player injuries or the margin of victory in previous games. In disagreement with a majority of literature since Pankoff
(1968), Woodland and Woodland find that the NFL betting market is highly efficient.
They find that the NFL betting market does not facilitate contrarian-betting strategies as well as the stock market does. The major result of the paper is that the NFL betting market isolates the effectiveness of contrarian strategies while holding risk levels of
Early Season Inefficiencies in the NFL Sports Betting Market 77
bettors constant. Woodland and Woodland’s finding that contrarian strategies are not profitable in NFL betting markets further shows how difficult it is for bettors to develop consistently profitable strategies in NFL betting markets.
Vergin (2001) studies the overreaction bias of bettors towards positive performance in previous games and the under reaction of bettors towards negative performance in previous games. Vergin contends that bettors prefer teams that have had recent strong positive performance over teams that have had recent strong negative performances. This bias is evident in the fact that over the 1969-1995 seasons, heavy favorites covered the spread only 46.6% of the time. Ultimately,
Vergin contends that this overreaction bias provides another example of inefficiency and refutes the previous findings that the Efficient Markets Hypothesis is present in the NFL sports betting market.
This paper proves to be the most similar to the bias we described in the attached papers. However, differences exist in the fact that Vergin (2001) focuses on biases that exist based on performance in the previous game, two to five games, or season. Our papers specifically examine two biases. One is a holdover bias that exists from season to season and manifests itself solely in the opening week of the season.
Bettors rely too heavily on the performance of teams in the prior season in judging the relative strength of teams. This bias allows for the formation of a profitable betting strategy the bets against teams that performed well in the prior season when playing against teams that performed poorly in the prior season. The second bias is related to betting on total score in the first week of the season. Statistically significant profits
Early Season Inefficiencies in the NFL Sports Betting Market 78
can be made by betting total scores of all games in the first week of the will fall below the over/under line. We show that this can be attributed to poor performance of offenses relative to defenses in the first week. Our tight focus on two betting strategies for the first week of the season led to statistically significant findings for both market inefficiencies and profitability, distinguishing our papers from the majority of the NFL sports betting market literature.
Dare and Holland (2004) find that previously documented inefficiencies of betting on home underdogs are not consistent from season to season. Dare and
Holland show that while some profits to betting on home underdogs appear in out-of- sample simulations, one cannot reject the hypothesis of market efficiency in favor of abnormal profits. They ultimately conclude that the expected profits arising from a bias favoring home underdogs may be too small to be exploited.
Burkey (2005) notes that authors who search for profitable trading rules which, ex post, prove to have been successful in earlier periods, will surely succeed if they investigate enough strategies. Burkey suggests the use of arbitrage when betting in
NFL markets. Burkey defines arbitrage as “buying and selling contracts in different markets that have either an expected or guaranteed profit.” Burkey contends that picking a strategy and obtaining an “advantage” with a local bookie’s line relative to the Vegas line is a profitable strategy for betting in NFL markets. Ultimately he argues that the essential element of a profitable betting strategy is not the strategy itself, but the size of the advantage obtained through finding different lines or odds in different markets. Burkey’s approach to exploiting the NFL market inefficiencies differs from
Early Season Inefficiencies in the NFL Sports Betting Market 79
the other literature through his focus on arbitrage rather than simply finding inefficiencies in the market.
Paul and Weinbach (2007) continued the search for profitable betting strategies in NFL sports betting markets by following the strategies of Levitt (2004) who shows sports books tend to set point spreads for National Football League (NFL) games to maximize profits, not to clear the market. Paul and Weinbach used data from www.sportsbook.com and found that the point-spread market in the 2006 NFL season was consistent with the Levitt’s hypothesis that betting dollars were not balanced.
According to the data, road favorites received a greater percentage of betting volume.
The data also showed that the percentage bet on the favorite became greater as the point spread increased. Paul and Weinbach therefore contend that by simply betting against the public when the sports book was unbalanced with 70% percent or more of betting volumes on the favorite, the bettor was able to earn positive returns.
Paul and Weinbach (2011) expanded upon their previous work by adding another season of data and by utilizing another source of betting percentage data,
Sports Insights. Paul and Weinbach (2011) affirm the findings of Paul and Weinbach
(2007) and Levit (2004) who show sports betting books are unbalanced. Paul and
Weinbach (2011) conclude that bettors tend to over bet good teams on the road and that the percentage of bets on the favorite increases with point spreads. Forming a betting strategy based on the unbalanced nature of NFL betting markets earns profits that are significant at the 10% level for the periods studied in both Paul and Weinbach
(2007) and Paul and Weinbach (2011).
Early Season Inefficiencies in the NFL Sports Betting Market 80
Paul and Weinbach (2011) adds two significant contributions to the literature on NFL sports betting markets. The first contribution is that of the affirmation of
Levitt (2004), which suggests NFL sports betting markets do not price to clear the market but rather to maximize profits. Secondly, Paul and Weinbach contend that NFL sports betting markets are not efficient because bettors share a preference for favorites and overs because of their inherent entertainment value over the alternative wagers.
Paul and Weinbach is in agreement with previous literature showing inefficiencies do exist in the NFL market, and that profitable betting strategies can be developed to exploit these inefficiencies.
Using data from all games from 1985-2010, roughly 5,876 games, Wever and
Aadland (2011) show that a differential strategy of betting on large underdogs can produce statistically significant profitable returns with a roughly 60% winning percentage. They contend that recently NFL betting markets have underpriced large underdogs and bettors have failed to recognize the increasing parity in the NFL. The findings of Weber and Aadland confirm those of Levitt (2004) and Paul and Weinbach
(2011) who contend that betting on large underdogs at home and even larger underdogs away can prove to be a profitable betting strategy.
As the presented literature contends, the thoughts surrounding the efficiency of the NFL sports betting market vary greatly on the level of the market’s efficiency as well as whether or not profitable betting strategies exist in markets. Our papers provide support that there are inefficiencies in the NFL sports betting market, namely in the first week of the season. This identified inefficiencies has yet to be documented
Early Season Inefficiencies in the NFL Sports Betting Market 81
and therefore stand as a significant addition to the literature on inefficiencies in NFL sports betting markets.
Additionally, our paper presents two significantly profitable strategies allowing bettors to take advantage of early season inefficiencies. Our previously undocumented strategy of betting against prior playoff teams in the first week of the season is one of the few betting strategies discussed in the literature that provides results that are significant at the 1% level. The strategy also provides an unheard of profit margin of
25.6% when applied over the 2004-2011 NFL seasons. None of the profitable strategies presented in the literature provides such a high profit margin. The strategy that bets under total lines in the first week of the season, also previously undocumented, provides statistically significant profits exceeding 13.6% per game on average. Our strategies are unique compared to strategies presented in past literature given their simple nature and high, statistically significant profit margins.
Our paper stands to be a significant addition to NFL sports betting literature due to our discovery of a new inefficiency in the market and development of a simple and profitable betting strategy.
References:
Early Season Inefficiencies in the NFL Sports Betting Market 82
Avery, C., & Chevalier, J. (1999). Identifying investor sentiment from price paths: the case of football betting. Journal of Business, 72(4), 493-521.
Badarinathi, R., & Kochman, L. (1996). Football betting and the efficient market hypothesis. The American Economist, 40(2), 52-55.
Barberis, N., Shleifer, A., & Vishny, R. (1998). A model of investor sentiment. Journal of Financial Economics, 49, 307-343.
Brown, G., & Cliff, M. (2005). Investor sentiment and asset valuation. Journal of Business, 78(2), 405-440.
Burkey, M. (2005). On “arbitrage” and market efficiency: An examination of NFL wagering. New York Economic Review, 36(1), 13-28.
Cooper, M., Dirmitrov, O., & Rau, P. R. (2002). A rose.com by any other name. Journal of Finance, 56(6), 2371-2388.
Dare, W. H., & Holland, A. S. (2004). Efficiency in the NFL betting market: modifying and consolidating research methods. Applied Economics , 36 (1), 9-15.
Edwards, W. (1968). Conservatism in human information processing. In: Kleinmutz, B. (Ed.), Formal Representation of Human Judgment, Wiley, New York.
Fisher, K., & Statman, M. (2000). Investor sentiment and stock returns. Financial Analysts Journal, 56(2), 16-23.
Gandar, J., Zuber, R., O'Brien, T., & Russo, B. (1988). Testing Rationality in the Point Spread Betting Market. The Journal of Finance , 43 (4), 995-1008.
Golec, J., & Tamarkin, M. (1991). The degree of inefficiency in the football betting market. Journal of Financial Economics , 30, 311-323.
Gray, P. K., & Gray, S. F. (1997). Testing Market Efficiency: Evidence from NFL Sports Betting Market. The Journal of Finance , 52 (4), 1725-1737.
Han, B. (2008). Investor sentiment and option prices. Review of Financial Studies, 21(1), 387-414.
Haruvy, E., Lahav, Y., & Noussair, C. (2007). Traders’ expectations in asset markets: Experimental evidence. The American Economic Review, 97(5), 1901-1920.
Krieger, K., Fodor, A., & Stevenson, G. (2011). The sensitivity of findings of expected bookmaker profitability. Journal of Sports Economics, forthcoming.
Early Season Inefficiencies in the NFL Sports Betting Market 83
Lacey, N. (1990). An estimation of market efficiency in the NFL point spread betting market. Applied Economics 22 (1), 117-129.
Lee, M., & Smith, G. (2002). Regression to the mean and football wagers. Journal of Behavioral Decision Making, 15(4), 329-342.
Levitt, S. (2004). Why are gambling markets organized so differently from financial markets? Economic Journal 114 (495), 223-246.
Millman, C. (2001). The odds: one season, three gamblers, and the death of their Las Vegas, Public Affairs (Perseus Books Group), New York, New York.
Neal, R., & Wheatley, S. (1998). Do measures of investor sentiment predict returns? Journal of Financial and Quantitative Analysis, 33, 523-547.
Pankoff, L. (1968). Market efficiency and football betting. Journal of Business, 41(2), 203-214.
Paul, R., & Weinbach, A. (2007). Does sportsbook.com set pointspreads to maximize profits? Tests of the Levitt model of sportsbook behavior. Journal of Prediction Markets, 1(3), 209-218.
Paul, R., & Weinbach, A. (2011). NFL bettor biases and price setting: further tests of the Levitt hypothesis of sportsbook behaviour. Applied Economics Letters, 18(2), 193- 197.
Sapra, S. (2008). Evidence of betting market intraseason efficiency and interseason overreaction to unexpected NFL team performance 1988-2006. Journal of Sports Economics, 9(5), 488-503.
Sauer, R. D., Brajer, V., Ferris, S. P., & Marr, M. W. (1986). Holding Your Bets: Another Look at the Efficiency of the Gambling Market for National Football League Games. Journal of Political Economy , 96 (1), 206-213.
Schmeling, M. (2009). Investor sentiment and stock returns: some international evidence. Journal of Empirical Finance, 16(3), 394-408.
Strumpf, K. (2003). Illegal sports bookmakers, University of North Carolina, Department of Economics.
Sturgeon, K. (1974). Guide to sports betting, Harper & Row, New York, 1974. Tryfos, P. P., Casey, S. S., Cook, S. S., Leger, G. G., & Pylypiak, B. B. (1984). The profitability of wagering on NFL games. Management Science, 30(1), 123-132.
Early Season Inefficiencies in the NFL Sports Betting Market 84
Vergin, R. C. (1998). The NFL pointspread market revisited: anomaly or statistical aberration. Applied Economics Letters, 5(3), 175-179.
Vergin, R. C. (2001). Overreaction in the NFL point spread market. Applied Financial Economics, 11(5), 497-509.
Vergin, R. C., & Scriabin, M. (1978). Winning strategies for wagering on National Football League games. Management Science , 24 (8), 809-818.
Wever, S., & Aadland, D. (2012). Herd Behaviour and underdogs in the NFL. Applied Economics Letters , 19 (1), 93-97.
Woodland, B., & Woodland, L. (2000). Testing contrarian strategies in the National Football League. Journal of Sports Economics, 1(2), 187-193.
Zuber, R. A., Gandar, J. M., & Bowers, B. D. (1985). Beating the Spread: Testing the Efficiency of the Gambling Market for National Football League Games. Journal of Political Economy , 93 (4), 800-806.