Believe the Hype

[ Pobierz całość w formacie PDF ]
Believe the Hype:
The Salience of National Rankings for College Sports Betting Markets and Implications for
Market Efficiency
May 12, 2004
Doug Geyser
Department of Economics
Stanford University, Stanford, CA 94305
e-mail:
doug.geyser@stanfordalumni.org
Advisor: Justin Wolfers
ABSTRACT
In this paper, I examine the salience of perceived expert opinion, in the form of national
rankings, for the betting behavior of collegiate basketball and football bettors. In collegiate
athletics, teams are ranked by certain national polls that are published widely but do not
necessarily reflect the true skill level of a given team. As such, the betting population may place
too much emphasis on the fact that a team boasts a ranking, thus causing the proportion of bets in
favor of that team to rise above the teamÓs objective probability of winning, and leading to a
violation of market efficiency. I find that national rankings are efficiently incorporated into the
point spread for college basketball; however, for college football, I find that bettors do not place
enough
emphasis on national rankings. Instead, my results suggest that a market inefficiency
exists in the college football betting market because the higher ranked team is significantly
more
likely to beat the spread when controlling for the favorite/longshot distinction, and thus expert
opinion does matter.
Keywords: sports betting, efficient markets, salience effects, expert opinion
______________________________________________________________________________
I am enormously indebted to my advisor Justin Wolfers for introducing me to this topic and for his consistent
patience and thoughtful comments throughout each stage of this process. His advice has improved the quality of this
thesis immeasurably. I would also like to thank my family and friends for their encouragement. All errors are my
own.
Doug Geyser
2
May 12, 2004
1. Introduction
In this paper, I concentrate on the salience of perceived expert opinion by exploring the
effect of teamsÓ national rankings on betting decisions in collegiate menÓs basketball and football
betting markets. Various sources widely publish national rankingsÏwhich can be considered
indexes of expert opinionÏyet these rankings do not necessarily reflect the true skill level of a
team. The two most popular polls are sponsored by the Associated Press (which I use), which
consists of sportswritersÓ rankings, and ESPN/USA Today, which consists of coachesÓ rankings.
If bettors place too much weight on the fact that a team is ranked, the proportion of bets in favor
of that team beating the spread may exceed the teamÓs objective probability of winning. Since
the bookmaker establishes the spread to account for this overbetting, such a bias in favor of
nationally ranked teams could lead to a profitable betting strategy and thus a market inefficiency.
In the popular view of typical basketball and football betting markets, the bookmaker or
betting institution theoretically establishes a spread that equalizes the amount of money bet on
each team. That is, if a majority of the dollars in play favor one team, the bookmaker changes the
spread to make the other team appear more attractive to bettors. The efficient markets hypothesis
then dictates that the point spread captures all available information regarding the gameÓs
outcomeÏin other words, no piece of information should serve as a predictor of whether a team
will beat the spread. Although this is the traditional explanation of the mechanics of spread-
setting, Levitt (2002) suggests that it may be inaccurate. He argues that not only do bookmakers
fail to equalize the amount of money bet on each team, but they Ðappear to be strategically
setting prices in order to exploit bettorsÓ biasesÑ (p.4). Regardless of which view holds true for
collegiate basketball and football betting markets, bookmakers still adjust the spread to account
Doug Geyser
3
May 12, 2004
for individualsÓ betting preferences. Since national rankings affect bettorsÓ decisions, they
consequently affect the spread in either situation.
If a bettor picks the underdog (favorite), he wins if the difference in actual points scored
is less (greater) than the spread. If the actual difference equals the spread, all bets are returned.
Unlike some other sports betting markets, the payoff described here does not depend on the
number of total players who picked the same team (i.e., it is not parimutuel). Instead, every
winning player receives $21 for every $11 bet. This Òeleven for tenÓ rule, as it is known, is
crucial because it guaranteesÏif equal amounts of money are bet on each sideÏthe bookmaker a
riskless profit by allowing him to retain the extra dollar as a commission, or Ðvigorish.Ñ The rule
consequently forces a bettor to win at least 52.4% of his bets to achieve a positive net profit
level. I hypothesize that a prudent bettor can use national rankings to guide betting strategy and
exploit the spread.
The underlying theory of this hypothesis is based on notions of psychology and
behavioral finance. Salience effects, which I discuss in more detail below, are considered in
psychology to be the overly influential effects on decision processes resulting from the increased
visibility or knowledge of a person or object that does not necessarily have a valid, logical
bearing on the decision. I suggest that the Associated Press (AP) poll is sufficiently visible to
unduly influence the decisions of bettors via a salience effect. In Hersh ShefrinÓs (2000)
excellent survey of behavioral finance, he thoroughly discusses availability and
representativeness heuristics, which arise prominently in the discussion of salience effects that I
will cover below. At least one other concept that he addresses is relevant to this paper. Shefrin
relates that Ðinvestors appear to cling to the idea that good stocks are the stocks of good
companies, and vice versaÑ (p.82). Thus, they invest simply in good companies, which may not
Doug Geyser
4
May 12, 2004
earn them the best returns. Another related idea from financial markets is the ÒS&P effectÓ (see,
e.g., Shleifer 1986, and Harris and Gurel 1986), the increase in a stockÓs price following its
addition to the Standard & PoorÓs 500 Index, where the addition acts as a Òcertificate of qualityÓ
for the stock (see, e.g., Dhillon and Johnson 1991, and Jain 1987). Similarly, I hypothesize that
the salient AP poll marks some teams as Ògood teams,Ó though betting on these teams may not
actually win bets in Las Vegas.
Collegiate sports betting markets offer a ripe opportunity for studying this question
because the number of teams is far greater than that in professional sports leagues. As a result, it
is more difficult for bettors to obtain information on every team, especially those teams
representing the smaller schools that are less likely to appear in the national polls. In situations
involving particularly small schools, it may be even more common for a bettor to overestimate
the ranked teamÓs chance of winning simply because it is nationally ranked. This topic therefore
is important because it relates to the extent to which decision-making is influenced by the advice
of specialists and by indexes that ostensibly denote quality, issues also prevalent in the study of
financial markets. That said, it should be noted that the strength of the analogy between sports
betting markets and stock markets is debatable, so one must be wary when making comparisons
(see Tetlock 2004 for a summary of the arguments for and against this analogy, and for an
analysis that argues that inefficiencies in sports betting markets do not generalize to financial
markets).
This paper differs from previous work in terms of both focus and empirics. First, though
other authors have studied college sports betting markets, to the best of my knowledge only one
previous study (Fair and Oster 2003) has concentrated on national rankings, which offer a unique
opportunity to make inferences concerning the importance and impact of expert opinion. I
Doug Geyser
5
May 12, 2004
examine the salience issue and game outcome relative to the spread, whereas Fair and Oster
instead primarily focus on the predictive power of the less publicized (and thus less salient)
computerized college football ranking systems for game outcomes only. Avery and Chevalier
(1999) also examine both expert opinion and team prestige, but only in their consequences for
movements in the point spreads for National Football League games, for which national rankings
do not exist and a good amount of information is readily available.
Second, I distinguish my methodology from that of Fair and Oster (2003), who use the
typical betting market regression in their sole point spread-related examination. This model is an
ordinary least squares (OLS) regression that places the difference in actual points scored as the
dependent variable and places the spread as the independent variable, sometimes including other
explanatory factors. I use two models as improvements over this OLS regression: a probit model
from Gray and Gray (1997) that regresses whether a team beat the spread on a variety of
explanatory variables, and a linear probability model (LPM) with the same variables. There are
several reasons to use the probit, on which I will elaborate below. But the best reason to eschew
the standard OLS regression model in favor of the discrete-choice probit model (and the LPM)
is, as Gray and Gray explain, Ðbecause the return on a bet is the same whether a team beats the
spread by 1 point or by 30 pointsÑ (p.1726). In other words, the only question that matters is: Did
the team beat the spread?
I use these models to investigate the salience of a national ranking in and of itself, as well
as combined with typical informative factors such as home team and favorite team indicators. I
find that national rankings hold no predictive significance for the college basketball betting
market and thus imply no violation of market efficiency. For college football, I conclude that
national rankings, when controlling for the favorite/longshot effect, are statistically significant
[ Pobierz całość w formacie PDF ]