Welcome to Scribd. Sign in or start your free trial to enjoy unlimited e-books, audiobooks & documents.Find out more
Download
Standard view
Full view
of .
Look up keyword
Like this
1Activity
0 of .
Results for:
No results containing your search query
P. 1
Points Having

Points Having

Ratings: (0)|Views: 4|Likes:
Published by baschance

More info:

Published by: baschance on Nov 26, 2008
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

06/14/2009

pdf

text

original

 
EXPOSING CHEATING AND CORRUPTION
Point Shaving: Corruption in NCAA Basketball
 By
J
USTIN
W
OLFERS
*
A new field of “forensic economics” has be-gun to emerge, applying price-theoretic modelsto uncover evidence of corruption in domainspreviously outside the purview of economists.By emphasizing the incentives that yield cor-ruption, these approaches also provide insightinto how to reduce such behavior. This papercontributes to this agenda, highlighting how thestructure of gambling on college basketballyields pay-offs to gamblers and players that areboth asymmetric and nonlinear, thereby en-couraging mutually beneficial effort manipula-tion through “point shaving.” Initial evidencesuggests that point shaving may be quitewidespread.The incentives for gambling-related corrup-tion derive from the structure of basketball bet-ting. To highlight a simple example, theUniversity of Pennsylvania played Harvard onMarch 5, 2005, and was widely expected to win.Rather than offering short odds on Penn win-ning the game, bookmakers offered an almosteven bet (bet $11 to win $10) on whether Pennwould win relative to a “spread.” In this exam-ple, the spread was
14.5, meaning that a beton Penn would win only if Penn won the gameby 15 or more points, while a bet on Harvardwould be successful if Harvard either won, orlost by 14 or fewer points.The incentive for corruption derives directlyfrom the asymmetric incentives of players, whocare about
winning
the game, and gamblers,who care about whether a team beats (or
covers
)the spread.Indeed, the example above is ripe for corrup-tion: the outcome that maximizes the joint sur-plus of the Penn players and the gambler occurswhen Penn wins the game, but fails to cover thespread (and the gambler has bet on Harvard).The contract required to induce this outcomesimply involves the gambler offering a contin-gent payment to the player, with the contin-gency being that he pays only if Penn fails tocover the spread. Given the player’s (approxi-mate) indifference over the size of the winningmargin, even small bribes may dominate hisdesire to increase the winning margin above 14points, and this, in turn, yields large profits forthe gambler who has bet accordingly. The bet-ting market offers a simple technology for thegambler to commit to paying this outcome-contingent bribe: he can simply give the playerthe ticket from a $1,000 bet on his opponent notcovering the spread.Such attempts to shave the winning marginbelow the point spread are colloquially referredto as “point shaving” and form the focus of myinquiry. I start by outlining the type of corrup-tion that theory suggests will be most prevalent:
Players will be bribed
not 
to cover the pointspread: It is easy for a player to reduce hiseffort in response to marginal financial incen-tives. By contrast, if he usually plays at closeto maximal effort—and if the point spread isset on this assumption—then inducing evengreater effort will be impossible, or expensive.
Favorites are more likely to shave points thanare underdogs: For an underdog to commitnot to cover the spread implies committing tolosing the game, while the favorite can bothwin and not cover. Thus, the payment re-quired to motivate the underdog not to coveris typically larger than that required to inducethe favorite to shave points.
 Discussants:
Raymond Fisman, Columbia University;Bruce Sacerdote, Dartmouth College.* The Wharton School, University of Pennsylvania,1456 Steinberg Hall-Dietrich Hall, 3620 Locust Walk, Phil-adelphia, PA 19104, and CEPR, IZA, and NBER (e-mail: jwolfers@nber.org). Thanks to Steven Levitt and BetseyStevenson for helpful feedback, and to Ravi Pillai, LisaRutherford, and Bryan Elliott for exceptional researchassistance.
279
 
Stronger favorites are more likely to shavepoints than weak favorites: For example,when the spread is three points, an attempt towin by only one to two points may backfire,either leading the player to lose the game(raising his cost of point shaving) or the gam-bler to lose his bet (lowering the benefit). Bycontrast, a team that shaves an expected 14-point victory to a margin of nine points facesvery little risk.
While point shaving may affect whether astrong favorite covers the spread, it shouldhave no effect (or minor effects) on theteam’s chances of winning the game. Thus,point shaving will lead us to observe “toomany” teams failing to cover, yet still win-ning the game.
Given the discontinuity in gamblers’ pay-offsat the spread, there may be a sharp differencein the probability of strong favorites just fail-ing to cover, compared to the proportion justcovering the spread.Further, point shaving is less likely when theprobability of detection is high, and more likelywhen the reward is large. Unfortunately, in thesports betting context, these two factors arelikely highly correlated, as televised games,high profile, and high-stakes games all offerlarger potential profits due to thicker bettingmarkets, but invite greater scrutiny and hencerisk of detection.
I. A Prima Facie Case
Data on the outcomes of 36,003 NCAA Di-vision I basketball games played between1989–1990 and 1996–1997 were provided byVegasInsider.com, and data for an additional37,775 games played between 1997–1998 and2004–2005 were extracted from Covers.com bya Web crawler. Bookmakers tended to take betsonly on more popular match-ups and, hence,only 60 percent of these games yielded usefulbetting data, for a final sample of 44,120 games.Figure 1 shows that the spread provides anunbiased forecast of game outcomes and ex-plains much of the variation. Note that the ef-ficient-markets hypothesis makes no predictionsabout whether the spread,
on average,
predictsthe winning margin, but, rather, suggests theprobability that a team beats the spread is un-predictable. Thus, the lower panel shows a morerelevant metric: the frequency with which thehome team covers, as a function of the spread.This panel yields little evidence of unexploitedprofit opportunities. Overall, the favorite beatthe spread in 50.01 percent of games. Strongfavorites (those favored to win by more than 12points;
n
9,244) covered only 48.37 percentof the time, a statistically significant deviationfrom semi-strong form efficiency. While thisevidence might be considered suggestive of point shaving, this implication follows only if the betting market systematically underesti-mates the incidence of point shaving.The more direct implications of the theoryconcern the full distribution of outcomes. Thetop panel of Figure 2 shows kernel densityestimates of the distribution of winning mar-gins—relative to the spread—for all games inthe sample (excluding strong favorites). If thespread is a prediction market-generated medianforecast (Wolfers and Eric Zitzewitz, 2004),then the deviation of game outcomes from thespread is a forecast error. These forecast errorsare roughly normally distributed with zeromean and a standard deviation of 10.9 points.More intriguingly, the bottom panel showsthe comparable distribution for strong favorites.Compared to the normal distribution, this figuresuggests that “too few” strong favorites beat the
F
IGURE
1. NCAA B
ASKETBALL
: G
AME
O
UTCOMES ANDTHE
S
PREAD
 Note:
Sample includes 44,120 NCAA Division I gamesfrom 1989 to 2005.
280 AEA PAPERS AND PROCEEDINGS MAY 200
 
spread and that this missing probability mass islargely displaced to outcomes in which the teamwins, but fails to cover. The left tail of thisdistribution largely follows the estimated nor-mal distribution, suggesting that while strongfavorites differ in their behavior with respect tocovering the spread, their likelihood of losingthe game remains largely unchanged. These ob-servations are consistent with some strong fa-vorites point shaving.To the extent that these inferences rely oncomparisons of an empirical distribution with aspecific parametric distribution, one might beconcerned that this analysis rests heavily onassumed functional forms. Simply comparingthe two empirical distributions to each otheralso yields similar conclusions, however.To estimate convincingly the overall preva-lence of point shaving and provide statisticaltests would require explaining the empirical dis-tribution of outcomes shown in Figure 2 as amixture of the distribution of outcomes whenpoint shaving and the distribution when notpoint shaving, weighted by their relative prev-alence. While we are directly interested in theseprevalence weights, the component distribu-tions are not directly observable, presenting anidentification problem. An explicit structuralmodel of player behavior is required to recoverthe distribution of outcomes when point shavingand this is the focus of Wolfers (2006).A simple shortcut yields some illustrativeestimates. Under the null of no point shaving,and an assumption that the distribution of fore-cast errors is symmetric,(1)
p
0
Winning margin
Spread 
 p
Spread 
Winning margin
2
Spread 
.Figure 3 shows these probabilities withineach point-spread decile. Naturally the propor-tion of teams that win, but fail to cover, rises asthe spread rises (as the range of outcomes thisincludes increases). Equally, as equation (1)suggests, the proportion of teams in the com-parison set of outcomes is typically just as large,as it covers just as wide a range of outcomes.There is a clear difference among strong favor-ites, however, with significant evidence of “toomany” teams winning but failing to cover, rel-ative to the comparison set.Among teams favored to win by 12 points ormore, 46.2 percent of teams won but did not
F
IGURE
2. P
ROBABILITY
D
ISTRIBUTION
: G
AME
O
UTCOMES
R
ELATIVE TO THE
S
PREAD
 Note:
Dashed line shows normal distribution, with meanzero. Solid line shows nonparametric estimated distribution.F
IGURE
3. NCAA B
ASKETBALL
: B
ETTING
L
INE AND
B
ET
O
UTCOMES
 Note:
Shaded area shows 95-percent confidence interval.
281VOL. 96 NO. 2 EXPOSING CHEATING AND CORRUPTIO

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->