Arbitrage Strategies For Cross-Track Betting On Major Horse Races

Arbitrage Strategies for Cross-Track Betting on Major Horse Races
Author(s): Donald B. Hausch and William T. Ziemba

Source: The Journal of Business, Vol. 63, No. 1, Part 1 (Jan., 1990), pp. 61-78
Published by: The University of Chicago Press
Stable URL: http://www.jstor.org/stable/2353237 .
Accessed: 23/06/2014 07:20
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to The
Journal of Business.
http://www.jstor.org
This content downloaded from 62.122.76.78 on Mon, 23 Jun 2014 07:20:17 AM

All use subject to JSTOR Terms and Conditions
Donald B. Hausch
University of Wisconsin-Madison
William T. Ziemba
University of British Columbia
Arbitrage Strategies for

Cross-Track Betting on Major
Horse Races*
I. Introduction Cross-track betting per-

mits bettors to place
Racetracks and securities markets have many wagers at their local
characteristicsin common. A difference,though, tracks on a race being
is their complexity; the racetrackis really a se- run at another track.
quence of markets that are relatively simple, Since each track oper-
ates a separate betting
short-lived,and, for the most part, independent. pool, the odds can vary
This "market-in-miniature"feature makes the across the tracks. The
racetrack attractive for tests of market effi- data suggest that the
ciency, especially since, as Thaler and Ziemba odds vary, and they of-
(1988, p. 162) suggest, "one can argue that wa- ten vary dramatically,
allowing arbitrage op-
geringmarketshave a better chance [thansecuri- portunities. This article
ties markets]of beingefficientbecause the condi- employs a risk-free ar-
tions (quick, repeatedfeedback) are those which bitrage model to dem-
usually facilitate learning." The many empirical onstrate the cross-track
racetrack studies support a weak form of ef- inefficiency and recom-
mends an optimal capi-
ficiency for some of the available wagers, while tal growth model for
other types of wagers seem not to be efficient. exploiting it. A simpler
These studies are reviewed in Section II. method is proposed for
This article studies cross-trackbetting, a rela- a single bettor at a sin-
tively new form of wagering.It allows bettors to gle cross track. The re-
sults indicate that these
wager at their track (a cross track) on a race methods would have
worked well in practice
* Withoutimplicatingthem, we would like to thankBruce on a number of recent
Faumanand Fraser Rawlinson.Also, we greatlyappreciate Triple Crown races.
the data supplied by a numberof U.S. racetracks,and we
wish to thankVictor Lespinassefor suggestingthe one-track
model.
(Journal of Business, 1990, vol. 63, no. 1, pt. 1)
K 1990 by The University of Chicago. All rights reserved.
0021-9398/90/6301-0004$01 .50
61

62 Journal of Business
TABLE 1 Home-Track and Cross-Track Betting, Kentucky Derby
Home-Track No. Cross Home-Track Cross-Track

Year Attendance Tracks Betting ($) Betting ($)
1982 141,009 ... 5,011,575 ...
1983 134,444 ... 5,546,977 ...
1984 126,453 24 5,420,787 13,521,146
1985 108,573 32 5,770,074 14,474,555
1986 123,819 56 6,165,119 19,776,332
1987 130,532 73 6,362,673 20,829,236
1988 137,694 93 7,427,389 24,449,058
being run at another track (the home track). Since cross-trackbetting

tends to be limitedto majorraces, it gives the racingpublican opportu-
nity to bet on some of the world's finest racehorses.This makes it very
popular with the public. Cross-trackwagering can lead to increased
attendanceand revenues at the cross tracksand add to the revenues of
the home track througha fee (usually 5% of the handle) paid by the
cross tracks. Thus, all the tracks can increase profits.'
Separatepools for each trackmeans the payoffs at the varioustracks
can differ.2 Due to the costs of arbitrage in this setting, market
efficiency across the tracks would, for practicalpurposes, allow some
differences across the various sets of track odds. Considerablediffer-
ences, however, would suggest the possibilityof a marketinefficiency.
The data demonstratethat considerabledifferences do occur. For ex-
ample, a $2.00 win ticket on Ferdinand,the winner of the 1986 Ken-
tucky Derby, paid from $13.20 at Fairplex in Pomona, California,to
$90.00 at EvangelineDowns in Lafayette, Louisiana.3Obviously, bet-
tors would have preferredtheir win bets on Ferdinandto be made at
1. For an example of this effect, considerthe home-trackand cross-trackbettingon

the KentuckyDerby(see table 1). The introductionin 1984of cross-trackbettingon this
race has greatly increasedtotal Derby wagering.At the same time, cross-trackbetting
seems to have had, at worst, only a minoreffect on home-trackwagering.The cross
tracks' revenues can increase also. For instance, Illinois set a one-day pari-mutuel
record on KentuckyDerby Day, 1987.Cross-trackbettingon the Derby accountedfor
$1,326,239of the $4,534,879wageredthat day in the state. Anotherexample is Calder
Race Course in Florida.They set all-timerevenueand attendancerecordson Kentucky
DerbyDay, 1985.Attendancewas 23,105,and $2,775,645was wagered,$562,453of it on
the Derby.
2. In some cases, all the wagersat the varioustracksare summed.Then, on the basis
of these summedvalues, identicalpayoffsare madeat all the tracks.This is often called
"intertrack"wageringand typicallythe tracksare withinone state. Intertrackwagering
will not be consideredhere.
3. These extreme payoffs are not just limitedto the smallertracks. Two large-track
examplesare Hollywood Park, where Ferdinandpaid $16.80, and WoodbineRacetrack
in Torontowhere he paid$79.60. Alysheba,the 1987KentuckyDerbywinner,paidfrom
$15.80 at Hollywood Park, Californiato $30.20 at Beulah Park,Ohio. WinningColors,
the 1988 derby winner, paid $7.40 at Pimlico Race Course, Maryland,and $10.40 at
Beulah Park.

Arbitrage Strategies 63
Evangeline Downs. The nature of the pari-mutuelbetting system re-

quires that if Ferdinand paid less at Fairplex than at Evangeline
Downs, then another horse, were it to have won, would have paid
more at Fairplex than at EvangelineDowns. Thus, if we are able both
to learn the odds and place our bets at various tracks, it appearsthat
significantarbitrageopportunitiesmay exist.
Section III develops a risk-freearbitragemodel to demonstratethis
cross-trackinefficiency.The optimalcapitalgrowthmodel is studiedin
Section IV on the general cross-trackproblemand, in Section V, on a
simplerone-trackproblem.These models are tested on data from sev-
eral recent Triple Crown races. A final discussion is in Section VI.
II. Efficiency of the Various Betting Markets

Among the possible wagers at the track are the so-called straightwa-
gers to win, place, and show. They pay off when one's horse is at least
first, second, or third, respectively. The "exotic" wagers include
quinellas(requiringone to name the firsttwo horses), exactors (requir-
ing the firsttwo horses in the correctorder),trifectas(requiringthe first
three horses in the correct order), and daily doubles (requiringthe
winners of two consecutive races). Tracks have also extended the
daily-double concept to picking the winners of three, four, six, and
even nine consecutive races. These are very low-probabilitybets that
can have tremendouspayoffs, and they are very popularwith the rac-
ing public. Before exotic wageringwas offered by the tracks, bettors
could use parlays and other combinationsof wagers to constructlow-
probability/high-payoffsituations. Rosett (1965) analyzed these pos-
sibilities and demonstratedthat, except for extreme long shots, the
bettors were rationalin the sense that a simplebet would not be madeif
a parlay with the same probabilityof success had a greater return.
Similarly, Ali (1973) showed that the returnon a daily double is not
significantlydifferentfrom the returnon what is an identicalwager, the
correspondingparlay of win bets.
Unlike typical casino games, where the odds are fixed, the odds at
the track are determinedby the relative amountsthe bettors wager on
the horses and by the track's transactionscosts (the track's take and
breakage).Thus, "prices" are determinedat the track much like they
are in securities markets. Let
n = the numberof horses in a race;

T = the numberof tracks accepting wagers on the race;
W = the total amount bet by the public at track t on horse i to
win (i= 1, ..., n and t = 1,. ..,T);
W = 1i = the win pool at track t; and
Q = track t's payback proportion(typically from .80 to .86).

Then the payoff per dollar bet to win on horse i at track t is
Q'W'/Wi if horse i wins,

0 otherwise.
To determine the place payoff let P' be the amount bet to place on
horse i at track t and let P' = X,P'be track t's place pool. The payoff
per dollar bet to place on horse i at track t is
J1+(Q'PI - P' - Pj')I(2P') if the first two horses are i andj 2

l 0 if horse i is not first or second.
Thus, the track keeps (1 - QI)Pt and the place bets on i and j are
repaid. The remainder,the losing bets minus the track take, is then
split evenly between those who bet on i and those who bet on j. The
share for i bettors is then divided on a per-dollar-betbasis. The place
payoff on i does not depend on whetheri was firstor second but it does
depend on which horsej was the other top finisher.In a similarfashion
the payoff per dollar bet to show on horse i at track t is
1 + (QtSt -S SJ - S5)IQ(3S') if the first three horses are i,j,

and k, (3)
0 if horse i is not at least third.
Here S' is the show bet on horse i by the publicat trackt, and S' = Jis'
is track t's show pool. It is possible that (QtSt - M - SJ - SW)I(3SM)is
less than 0.05, or even negative. In these cases, called minuspools, the
track usually agrees to pay $0.05 profitfor each dollarwagered. Minus
pools can also occur in the win and place markets, but are much less
common. Equations(I)-(3) ignore breakage,the additionalchargethat
results from the track roundingall payoffs down to the nearest 5 or 10
cents on the dollar.4
If prices reflect all available informationthen a marketis said to be
efficient (see Fama 1970).There are two conclusions that can be drawn
from the many studies of win marketefficiency (see, for instance, Ali
[1977];and Snyder [1978]).First, the North Americanpublicunderbets
favorites and overbets longshots, and this bias appears across the
many years that data have been collected and across all sizes of race-
trackbettingpools.5 Second, despite its strengthand stability,this bias
is almost always less thanthe tracktake and thus it cannotbe exploited
to achieve positive profits. Figure 1 illustrates the favorite/longshot
4. Breakagemay seem like a relativelyminorcost, but we (Ziembaand Hausch 1987)

demonstratethat it can have a dramaticlong-runeffect on a bettor'sfortune.
5. Busche and Hall (1988)demonstratean opposite bias for Hong Kong bettors.Con-
traryto the North Americanbettors, Hong Kong bettors tend to overbet favoritesand
underbetlong shots-a bias that is consistent with risk aversion.

1.2
1.0
________________________ Track Payback

= 8467
08
Expected
Return per
Dollar Bet
06
04
02
0 l l l l l l
1/10 2/5 1/1 2/1 3/1 5/1 10/1 20/1 30/1 50/1 100/1
Track Odds (Log Scale)
FIG. 1.-Expected return per dollar bet versus odds level: aggregationof
studiesinvolvingmorethan50,000races. (Source.-Ziemba andHausch 1986.)
bias and shows that the track take (assumed to be 15.33%, as it is in

California) is sufficiently large to preclude profits. The one profitable
exception is extreme favorites at odds of 3-10 or less. However, they
are relatively rare. Thus, if we define a market to be weakly efficient
(see Fama 1970), if no one can devise a profitable trading rule based on
historical price information, the win market is, for practical purposes,
weakly efficient.
Hausch, Ziemba, and Rubinstein (1981) tested the efficiency of the
place-and-show markets. They assumed:
ASSUMPTION 1. If qi is the probability that horse i wins, then the
probability that i is first andj is second is qiqj/(l - qi), and the probabil-
ity that i is first, j is second, and k is third is
qiqjqkl[(1 - qi)][( 1 -qi -qj)]
(These formulas were developed and tested by Harville [1973].)6
6. Henery (1981) and Stern (1987) show that these equationscan be derived by as-
sociating with each horse an independentexponential random variable with a scale

ASSUMPTION
2. The win market is efficient, so the win odds can be
used to estimate qi.
Using these two assumptions and equations (2) and (3), Hausch,
Ziemba, and Rubinstein(1981) were able to identify horses that were
underbet to place or show. The optimal capital growth model then
determinedthe place-and-show wagers that maximized the expected
rate of growth of one's bankroll. Since the exact model is a com-
plicated nonlinearoptimizationproblemthat is difficultto solve at the
track, Hausch, Ziemba, and Rubinstein (1981) developed simple re-
gression approximationswith quite minimaldata-entryrequirements.
Their empirical studies on two seasons of racing data indicated that
significantreturnson the order of 11%were possible in the place-and-
show markets. We (Hausch and Ziemba 1985; Ziemba and Hausch
1987) extended Hausch, Ziemba, and Rubinstein's (1981) results to
provide further evidence of the place-and-show inefficiency. We
(Ziembaand Hausch 1986)and Asch and Quandt(1987)studied ineffi-
ciency of the exotic markets. Asch, Malkiel, and Quandt(1984, 1986)
investigatedwhethera drop in the odds late in the bettingperiodmight
reflect inside informationand thereby point to wagers that may have
positive expected returns. Their results suggest that this is not the
case, however. A more thorough literature survey is in Thaler and
Ziemba (1988) and Hausch and Ziemba (1990). The latter also studies
racing outside of North America.
III. Inefficiency of the Win Market and the Risk-free Hedging Model
The literaturehas demonstratedweak efficiencyof the win marketat a
single track, despite a favorite/longshotbias. To test whetherthis weak
efficiency is maintainedacross the win marketswith cross-trackbet-
ting, data were collected on several recent Triple Crown races. Al-
though cross-track betting is becoming more popular, it tends to be
restricted to major races. The best known of these are the Triple
Crown races: the Kentucky Derby at ChurchillDowns on the first
Saturday in May, the Preakness Stakes at Pimlico Race Course 2
weeks later, and the Belmont Stakes at Belmont Park 3 weeks after
that.
A simple problem is considered first. Allowing only win betting,
what is the minimumamountthat our bettor must wager to ensure the
returnof $1.00 regardlessof which horse wins the race?The solutionto
this problem is, for each horse, to identify the track that has it at the
longest odds and bet just enough to receive $1.00 if it wins. The solu-
tion involves no estimation of the horses' win probabilities. If this
parameterequal to the inverse of its win probability.Then, any orderingof the random
variablesis just the Harvilleformulas.Sternalso develops alternativeorderingformulas
using gammadistributionsthat are more accuratebut more complicated.

TABLE 2 Projected Win Payoffs, 1983 Preakness
$ Amount of
Highest Win Return Wager That
Horse No. (on a $1 Bet) Track Will Return $1
1 29.40 Louisiana Downs .0340

3 34.60 Los Alamitos .0289
4 169.90 Hollywood .0059
7 10.60 Pimlico .0943
9 116.10 Hollywood .0086
Total .9356
solutionwere employedat one track, then our bettorwould have to pay

1/Q dollars (an amountgreaterthan $1.00) to ensure a returnof $1.00.
With the opportunityof bettingat several tracks, each with a different
set of odds, we may be able to lower this minimumamountto below
$1.00. To see how this system works, considerthe 1983Preakness.The
final win odds were collected from 11 tracks that allowed wageringon
this race.7 The highest of the 11 win payoffs on horse number 1 was
$29.40 per dollar wagered at Louisiana Downs. Thus, a win bet of
$0.0340 there would have returned$1.00 when, in fact, horse number
1, Deputed Testamony, won the race. This wager and those on the
other horses are presented in table 2.
Thus, by wagering$0.9356, our bettoris guaranteed$1.00 regardless
of who wins the race. This is a certain profit of $0.0644 per $0.9356
wagered, or a guaranteedreturnof 6.9% rate of returnin a 2-minute
race.
Obviously, "risk-free" arbitrageis not possible. Implementingthe
system requiresthat the trackodds be sent to a centraldecision maker
several minutes before the end of betting,and duringthattime the odds
can change. The system does, however, demonstrate the large dis-
crepancies in bettingacross the tracks and shows how simple it can be
to take advantageof them. Table 3 presents the results of applyingthe
system to the data from other Triple Crown races.
Three of the races had insufficientvariance in the win odds across
7. The numberof outlets involvedin cross-trackbettinghas increasedover the years.

The 1985KentuckyDerbyhad bettingat 32 outlets, includingNew YorkCity's off-track
betting(OTB). The 1985Preaknessand Belmonthad28 and41 outlets, respectively.For
1988,the numberof outlets for these three races were, respectively,93, 87, and 76. We
requestedthe finalwin, place, and show datafromthe outlets thatwe knewhadaccepted
wagers. Response rates tended to be low and, in some cases, the data were no longer
available,havingbeen stored for only a short period.All the data receivedwere used in
the analysis.

TABLE 3 Risk-free Hedging Model Results
No. Horses No.

Race in Race Tracks % Profit
Preakness:
1982 7 5 .0
1983 11 11 6.9
1984 10 4 .0
1985 11 7 2.5
Belmont:
1982 11 5 13.6
1983 12 9 8.5
1984 11 2 .0
1985 9 11 5.0
Kentucky Derby:
1984* 12 7 .1
1985 12 6 10.1
Average 4.7
* Cross-trackbettingon the KentuckyDerbydidnot beginuntil
1984.
the tracks to allow a risk-freeprofit.This is not surprisingfor the 1984

Belmont because we had data from only two tracks. Also, the results
for the 1982and 1984Preaknessare based on only five and four tracks,
respectively. In these three races our bettor would obviously make no
wagers. These three races giving a 0% returntogether with the other
seven races have an average risk-freeprofitof 4.7%. This rate of profit
is for the certain return of $1.00. As the requiredcertain return in-
creases, then the wagers begin to affect the odds and this rate of profit
will decrease.8
8. This "certainreturn"scheme can be extendedto includeplace-and-showbetting.

Let R' be the returnon a $1.00 win bet on horse i at trackt when i wins. Similarly,let R,'
be the returnon a $1.00place bet on i at trackt wheni andj are the top two finishers,and
let Rb'kbe the returnon a $1.00 show bet on i at trackt when i, j, and k are the top three
finishers.Formulas.()-(3), correctedfor breakage,determineRf, RMand R',k.The deci-
sion variablesare amountsto wagerto win, place, and show on horse i at trackt andare
represented,respectively,by w', pf, and s'. There is a constraintfor each ijkfinishthat
requiresa returnof $1.00, shouldthatfinishoccur. The followingformulationignoresour
effect on the odds and, while this effect is negligiblefor a returnof $1.00, it should be
includedfor largerrequiredreturns.The formulationthat determinesthe minimumex-
penditurefor a certainreturnof at least $1.00 is
I* ,,
minimize E I (wf + P! + sr),

subjectto
7'
E (RVw' + Rj,p! + Rj'ipj'+ Rb*ks + RMOks + Rkijsk) I for each i, j, k,
wf, Pi' s' for all i = 1, .

0 n and t= 1,. T.
This linear program has a large number of constraints; with n horses there are
n(n - 1)(n - 2) possible ijk finishes and, consequently, n(n - 1)(n - 2) constraints.(The
dual programhas 3Tn constraints,however.) Interestingly,even with the possibilityof
place-and-showbetting, betting only to win, as in Sec. III, may be optimal.The 1982

IV. The Optimal Capital Growth Model

Section III demonstrated an inefficiency in cross-track betting. We
now propose and test an "optimal" wagering strategy for this
inefficiency. Bettors at the racetrack appear to have many different
objectives and, therefore, employ many differentwageringstrategies.
Two quite reasonable long-termobjectives, however, are (1) to max-
imize the expected rate of growthof one's bankroll,and (2) to minimize
the expected time to reach some specified large wealth level. Breiman
(1961)proved that both of these objectives are asymptoticallysatisfied
by maximizing,on a myopic period-by-periodbasis, the expected log-
arithmof one's final wealth. This approachis termedthe optimalcapi-
tal growth model, and its theoreticaljustification for the logarithmic
utility function has been well studied (see Ziembaand Vickson [1975]
for references and a discussion of its assumptionsand results). A simu-
lation by us (Ziembaand Hausch 1986)suggests that this strategy not
only performswell asymptoticallybut over a year of wageringit can
also be expected to outperformother commonly used betting strate-
gies. Other attractive features of the capital growth model are that
one's effect on the odds can be accountedfor and bet size is monotone
in wealth. On the negative side, though, the recommendedbets can be
very large. Indeed, the Arrow-Prattabsolute risk-aversion index is
wealth- ', which is close to zero for large wealth.
Let w0 be our bettor's initial wealth and qijk the probabilitythat i is
first, is second, and k is third. Also, let P', = Pi + P5and S,k= S + Si
+ S5'.The optimal capital growth model is9
n n n_
max > > > qijk log wo

{W19p ,S'}
i= 1 ji.-l k=i?-l
+ f [Q'(W' +1 wI) - (Wf + w!)J( Wj+ )

n
+ + P( + f=1 ) - (Pg, + Pi + PD
Preakness(with 7 x 6 x 5 = 210 constraints)is one exampleof this. The reasonfor no

place-or-showbettingat optimumseems to be a coordinationproblem.Win bets return
one and only one positive amount and these payoffs are mutuallyexclusive. Thus, a
returnof $1.00 can be efficientlyguaranteedwith win bets. Place-and-showbets, how-
ever, tend to have many differentpossible payoffs, and collectingon them is not mutu-
ally exclusive. These differencesmake it difficultto efficientlychoose place-and-show
bets over the win bets.
9. It is very simple to include exotic wageringin this capital growth model. It was
formulatedto consideronly win, place, and show betting,though,because that was the
only availabledata for testing the model.

X-P, + P, PJ + PJ.
n
[Q(St + S) (Sik + s, )
- ? k+
3
X I( + +
k
n n n
(, e ,P + ESe)
f7lsi esi,j e?i,j,k _J

subject to
T n
>I >I (we + Pe + St) < WO,
We 09? PI 0' t =1, . . .,9 T, e = 1, .. n.

se
SI 0O
This model requiresestimates of the win probabilitiesfor each horse.

The efficiency studies of the win markethave indicatedthat the pub-
lic's win odds, adjusted for the favorite/longshotbias, provide good
estimates of these probabilities. However, with cross-trackwagering
we not only have a differentset of win odds for each participatingtrack
but, as demonstratedby the Ferdinandexample in Section I and the
risk-freehedging model in Section III, these odds can vary consider-
ably. Rather than take a weighted average of all the tracks' odds to
arrive at a set of win probabilities,we decided to use only the home
track's odds.'0 This decision was based on a perceived informational
advantagethat the home-trackpublic has over the bettors at the cross
tracks, an advantagethat results from several factors: (1) since these
races were run near the end of the day's racing, the home-trackpublic
had watched the jockeys perform in, perhaps, several races already,
they had observed the condition of the track and possibly noted any
track biases, and they saw the horses in the paddockand in the parade
to post; (2) the home crowd knows better if their track tends to favor
front-runnersor late chargers;and (3) since the home trackis usuallya
largertrack that has many majorraces, its publicis more likely to have
seen some of these horses race earlier in the season. Further, the
studies supportingwin-marketefficiency have all involved home-track
odds. Thus, we assume win-marketefficiency at the home track and,
10. This decision was not the resultof any analysisof the data. There are only a few
cross-trackraces each year and the requireddata for each race are the finaltote-board
figuresfromseveralof the racetrackspermittingthis betting.These difficultiesled to data
on only 10 races being collected, much less than would be requiredfor any analysis.

TABLE 4 1982 Preakness
Horse Finish Pimlico's Win Odds Win Probability
1. Reinvested ... 7.6-1 .090

2. Cut Away 3d 41.6-1 .011
3. Water Bank ... 12.0-1 .056
4. Bold Style ... 26.2-1 .023
5. Laser Light ... 5.3-1 .125
6. Linkage 2d .5-1 .597
7. Aloma's Ruler 1st 6.9-1 .098
after adjustingfor the favorite/longshotbias in figure 1, use the home-

win odds to estimate win probabilities.Since this win-marketefficiency
is not necessarily maintainedat the cross tracks, the model can then
include possible win betting at the cross tracks. The 1982 Preakness
Stakes will be used to illustrate this model. Table 4 lists the race's
entrants and their odds.
Aloma's Ruler's 6.9-1 win odds meant that, when he won, each
dollar wagered on him was returnedwith an additional$6.90 profit.
Linkagewas the crowd's favorite. His 0.5-1 win odds meantthat, had
he won, he would have returned$1.50 per dollarwageredon him. After
adjustingthe win probabilitiesfor the favorite/longshotbias, the proba-
bilities were then normalizedto sum to one. These probabilitiesand
equation(1) adjustedfor breakageallow the calculationof the expected
returnon an additionaldollarbet to win for each horse i at each trackt.
Similarly,the Harvilleformulasin assumption 1 and equations(2) and
(3), also adjusted for breakage, allow the calculation of the expected
returnson place-and-showbets (see Hausch, Ziemba, and Rubinstein
1981).We received final win, place, and show figuresfrom four of the
tracks that allowed betting on the 1982 Preakness: Golden Gate and
Los Alamitos in California,Centennialin Colorado,and Penn National
in Pennsylvania.The calculatedexpected returnson the variousbets at
the four tracks are presented in table 5.
If a track has a payback proportionof, say, 0.85 then the average
returnon $1.00 bets, ignoringbreakage,will be $0.85. Withbreakage,it
will be somewhat less than $0.85. For individualhorses the expected
returnsfrom table 5 vary from 0.181 (show bet on horse 2 at Golden
Gate) to 1.336 (place bet on horse 6 at Los Alamitos). Of particular
interest are the expected returns exceeding 1.00 since those indicate
the wagers that have a positive expected profit. We restricted our
attentionto wagers with expected returnsof at least 1.10. They are the
highlightedexpected returnsin table 5. If the probabilityestimates of
this model were exact, then this 1.10 cutoff could result in suboptimal
wagering,particularlysince diversificationcan even lead to the inclu-
sion of wagers with negative expected profits. Hausch, Ziemba, and

TABLE 5 Expected Returns, Cross-Track Betting on 1982 Preakness
Horse
1 2 3 4 5 6 7
Finish ... 3 ... ... ... 2 1

Win probability .090 .011 .056 .023 .125 .597 .098
Expected return on
a $1 bet to win:
Golden Gate .837 .246 .437 .708 .788 .955 1.000
Centennial .900 .389 .370 .570 .825 .836 1.147*
Los Alamitos .666 .391 .347 .777 .713 1.015 1.196*
Penn National .954 .484 .588 1.109* .650 .896 .755
Expected return on
a $1 bet to place:
Golden Gate .749 .193 .349 .582 .669 1.149* .880
Centennial .769 .238 .260 .329 .719 1.084 1.120*
Los Alamitos .794 .233 .277 .556 .778 1.336* .888
Penn National .673 .399 .391 .737 .586 1.101* .731
Expected return on
a $1 bet to show:
Golden Gate .837 .181 .405 .413 .817 1.153* .996
Centennial .747 .197 .340 .252 .803 1.008 1.138*
Los Alamitos .890 .200 .392 .341 1.180* 1.293* .873
Penn National .710 .235 .451 .388 .769 1.099 .793
NOTE.-In this table, horses2, 6, and7 are Cut Away, Linkage,and Aloma'sRuler,respectively.
*Highlightedexpected returns(returnsof at least 1.10).
Rubinstein (1981) found, however, that due to the approximationsin

the model it was more profitableto wager only if the expected return
was well in excess of 1.00. For large tracks and races of the qualityof
these Triple Crown races, a minimumexpected returnof 1.10 is rea-
sonable.
The capital growth model was run on the data from these cross
tracks. Table 6 presents the optimal portfolio of wagers assuming a
1.10 expected returncutoff and an initial wealth of $2,500. This port-
folio has wagers totaling $2,437 at all four tracks and its certainty
equivalent can be calculated as $432. Since the first three finishersof
this race were horses 7, 6, and 2, our bettor would collect on the win,
place, and show bets on horse 7 and the place and show bets on horse
6, for a returnof $3,716.90and a profitof $1,279.90.The payoffs in this
table account for our bettor's effect on the odds. For example, the
actualpayoff per $1.00 bet to show on horse 7 at Centennialwas $2.70.
However, our bettor's $83 show bet at this track would have lowered
this payoff to $2.50. The latter payoff was used to determineour bet-
tor's return.
Table 5 supposes only $1.00 is wagered.Thus, even thoughone track
may have a higher expected returnto, say, show on some horse than
another track has, it can be that our bettor will make show wagers at

TABLE 6 Optimal Capital Growth Wagers, Cross-Track Betting on 1982

Preakness
Expected $ Payoff $ Total

Horse Bet Track Return $ Bet per $1 Bet Return $ Profit
4 Win P.N. 1.109 14 ... ... - 14.00

7 Win L.A. 1.196 40 12.00 480.00 440.00
6 Place L.A. 1.336 855 1.50 1282.50 427.50
7 Place Cen. 1.120 46 3.30 151.80 105.80
S Show L.A. 1.180 172 ... ... - 172.00
6 Show G.G. 1.153 571 1.30 742.30 171.30
6 Show L.A. 1.293 656 1.30 852.80 196.80
7 Show Cen. 1.138 83 2.50 207.50 124.50
Totals 2,437 3,716.90 1,279.90
NOTE.-P.N. = Penn National; L.A. = Los Alamitos; Cen. = Centennial; G.G. = Golden Gate.
both tracksbecause of the bettor's effect on the odds. This happenedat

Golden Gate and Los Alamitos. Horse 6 had expected returnsof 1.153
and 1.293at these tracksand the optimalshow bets were $571and $656
at them, respectively. Withthese wagers, it must be thatthe additionto
expected utility from an additionaldollarbet to show will be the same
at the two tracks."l
The capital growth model was studied with cross-trackdata on the
other Triple Crown races. In each case it was assumed that the initial
wealth was $2,500. The results of these races and the 1982Preakness
are given in table 7.
Despite losses on two of the races, includinga huge loss on the 1984
Preakness, there was a total profit of $2,647.80 for a 15%returnon
money wagered. Average profits were $264.78. The standarderrorof
the mean is $342, so no statisticallysignificantstatementscan be made
about positive profits on the basis of these results. Studying the ex-
pected returns suggests that many of the profitableoverlays occur
because of regionalbiases. For instance, ConquistadorCielo, the win-
ner of the 1982 Belmont Stakes, raced his entire career on the East
Coast. The West Coast bettors were less familiarwith him, while on
the East Coast he was, for many, a sentimentalfavorite. In the Bel-
mont Stakes he tended to be sent off at lower odds at the East Coast
tracks. When he won he paid from $5.80 to win at CommodoreDowns
in Pennsylvaniato $15.40 at Los Alamitos in California.Another ex-
ample is Tolomeo, the English colt that won the 1983 ArlingtonMil-
lion. His odds were 4-1 in England and 38-1 at Arlington Park in
11. This marginalutility can be complicatedbecause a componentof the expected

returnon an additionaldollarbet to show on horse6 at Los Alamitosis the positiveeffect
that it has on the returnto the show bet on horse 5 at Los Alamitosif 6 finishesout of the
money.

0 cl C)
_4 0 (7-
C4 e'#0I0 C4
oN o
CZ r.ooh^N\>C
0 _
C) "t W) \_CD
6
z
ev 0_CZ o NC
.C o0 VI)t 0o o- 'r 0o
a- 'r1a r-r 00 r
CZ
~ C
o 6
zt>0 .z ^NFN O?r -
t 3 ~t ^ ^ ^ t ^ ^ ^0
0 r C) H
Lo
C:
Co> o* n W) _

Illinois. This observation and table 7 suggest that a high expected

returnis possible even with only a few carefullyselected cross tracks.
The Triple Crown races are the most widely publicizedof the North
American horse races. Thus, one might expect that other less pub-
licized cross-trackraces would have even more divergencein the odds
across the tracks. If so, they should have an even greaterpotentialfor
expected profits.
A majorassumptionof this model is that after all other bettors have
made their wagers our bettor (1) learns the tote-boardinformationat
each track, (2) runs the capital growth model, and (3) communicates
the optimalwagers to agents at each track, who then make the wagers.
Obviously this is extremely difficult, if not impossible, in practice,
never mind any legal concerns. Just learningthe tote-boardinforma-
tion at each trackis difficultbecause there are no pay phones inside the
racetrackgrounds. A central decision makerand the agents communi-
cating with cellular phones is feasible but still requires a significant
amount of time. Unfortunately, though, odds may change in the last
few minutesof betting, and profitablebets a few minutesbeforethe end
of betting may not be profitable based on the final odds. Hausch,
Ziemba, and Rubinstein(1981) studied the odds changes in the last 2
minutes of betting and found that expected returnsdid change some-
what but profitableplace-and-showbets 2 minutesfrom the end tended
to remain profitablebased on final odds. However, the agents in this
model, because of its extra complications, would probably have to
report odds more prematurethan those 2 minutes before the end of
betting. The results given in this article, then, may overestimate the
profits possible in practice. Additionally,our profitfiguresdo not ac-
count for the costs of implementingthis procedure-the agents and
other costs at each cross track, the long-distancephone calls, and the
computer time.
V. Testing the One-Track Capital Growth Model

Implementingthe previous section's capital growth model is certainly
difficult. However, when the race is televised from the home track
there are simplerversions of this scheme possible for one bettorat one
track. Our bettor, with a portable television at the cross track, can
view the home odds when they are shown on TV. With these odds
giving "true" win probabilities,our bettor can search for overlays at
the cross track. A flat bet to win could be made on horses going off at
longerodds than at the home track, or a sophisticatedbettorcould also
bringa portablecomputerto the trackand runthe capitalgrowthmodel
describedin Section IV for one track, that is, T = 1. This latterscheme
is tested here with the Triple Crown race data.
As an example, consider the 1984 Kentucky Derby simulcast at

TABLE 8 Optimal Wagers Based on One-Track Capital Growth Model, 1984

Kentucky Derby
Wager Expected $ Payoff on $ Realized

Type Horse Return $ Optimal Bet a $1 Bet Return $ Profit
Win 2 1.236 15 ... ... - 15.00

Win 10 1.159 ... 5.60 ... ...
Place 2 1.153 ... ... ... ...
Place 10 1.410 249 3.70 921.30 672.30
Show 2 1.353 261 ... ... -261.00
Show 10 1.275 295 2.50 737.50 442.50
Totals 820 1,658.80 838.80
Golden Gate Fields in Albany, California. Using the win odds televised
from Churchill Downs, Golden Gate had six wagers with expected
returns exceeding 1.10. With a wealth of $2,500, the capital growth
model yields the optimal wagers shown in table 8.
This portfolio has four wagers totaling $820. The win bet on horse
number 10 and the place bet on horse number 2 are zero even though
they have expected returns exceeding 1.10. This is because the possi-
bility of number 2 doing well in the race is better accounted for with the
higher returning win-and-show bets on him. Also, the possibility of
number 10 doing well is better accounted for with the higher returning
place-and-show bets on him. Swale, number 10, won the 1984 Derby,
followed by Coax Me Chad and At The Threshold for a 10-12-9 finish.
Therefore, only the place-and-show bets on Swale paid off for a return
of $1,658.80 and a profit of $838.80.
Table 9 presents the results of this one-track model on the other
races. The average wagers on a race varied from $78.33 to $1,173.25
and the average profits varied from - $868.67 to $824.10. The average
of these 10 average profits was $69.97 or 9.2% on the money wagered.
Again, there is such variability in the profits that, without additional
data, no statistically significant statements can be made about positive
expected profits.
VI. Final Discussion

There is considerable evidence that the win market at the racetrack is
weakly efficient. A risk-free arbitrage model is presented that demon-
strates that this is not the case with cross-track wagering. The mispric-
ing by the public at the cross tracks may be due partly to their more
limited access to information relative to those attending the home
track. Also, some of the variance in the odds across the tracks may be
due to certain horses being more familiar to bettors in certain regions of
North America. The optimal capital growth model suggests that the

TABLE 9 Results of One-Track Capital Growth Model on Triple Crown Races
Average Average Average

No. Cross-Track Cross-Track Cross-Track
of Cross Wager Realized Return Profit
Race Tracks ($) ($) ($)
Preakness:
1982 4 1,173.25 1,724.80 551.55
1983 8 909.25 353.79 - 555.46
1984 3 868.67 .00 - 868.67
1985 6 799.67 789.37 - 10.30
Belmont:
1982 4 407.25 639.02 231.78
1983 7 532.71 488.63 -44.08
1984 2 1,158.00 1,982.10 824.10
1985 9 78.33 7.80 - 70.53
Kentucky Derby:
1984 6 520.67 884.92 364.25
1985 6 1,161.33 1,438.42 277.08
Average 5.5 760.91 830.88 69.97
discrepancies across the tracks can allow profits, but further work is
needed to demonstrate significant profits. A simpler version of the
optimal capital growth model for one bettor at one cross track also
demonstrates the possibility of profit.
References
Ali, M. M. 1977. Probability and utility estimates for racetrack bettors. Journal of Polit-
ical Economy 85 (August): 803-15.
Ali, M. M. 1979. Some evidence of the efficiency of a speculative model. Econometrica
47 (March):387-92.
Asch, P.; Malkiel, B. G.; and Quandt, R. E. 1984. Market efficiency in racetrack betting.
Journal of Business 57 (April): 165-75.
Asch, P.; Malkiel, B. G.; and Quandt, R. E. 1986. Market efficiency in racetrack betting:
Further evidence and a correction. Journal of Business 59 (January): 157-60.
Asch, P., and Quandt, R. E. 1987. Efficiency and profitability in exotic bets. Economica
59 (August): 289-98.
Breiman, L. 1961. Optimal gambling systems for favorable games. In Proceedings of the
Fourth Berkeley Symposium, pp. 65-68. Berkeley: University of California Press.
Busche, K., and Hall, C. D. 1988. An exception to the risk preference anomaly. Journal
of Business 61 (July): 337-46.
Fama, E. F. 1970. Efficient capital markets: A review of theory and empirical work.
Journal of Finance 25 (May): 383-417.
Harville, D. A. 1973. Assigning probabilities to the outcomes of multi-entry competi-
tions. Journal of the American Statistical Association 68 (June): 312-16.
Hausch, D, B., and Ziemba, W. T. 1985. Transactions costs, extent of inefficiencies,
entries, and multiple wagers in a racetrack betting model. Management Science 31
(April)- 381-94.
Hausch, D. B., and Ziemba, W. T. (eds.). 1990, in press. Efficiency of Racetrack Betting
Markets. New York: Academic Press.

Hausch, D. B.; Ziemba, W. T.; and Rubinstein, M. 1981. Efficiency of the market for
racetrack betting. Management Science 27 (December): 1435-52.
Henery, R. J. 1981. Permutation probabilities as models for horse races. Journal of the
Royal Statistical Society, ser. B, 43:86-91.
Rosett, R. H. 1965. Gambling and rationality. Journal of Political Economy 73 (Decem-
ber): 595-607.
Snyder, W. W. 1978. Horse racing: Testing the efficient markets model. Journal of
Finance 33 (September): 1109-18.
Stern, H. S. 1987. Gamma processes, paired comparisons and ranking. Ph.D. disserta-
tion. Stanford, Calif.: Stanford University, Department of Statistics.
Thaler, R., and Ziemba, W. T. 1988. Parimutuel betting markets: Racetracks and lot-
teries. Journal of Economic Perspectives 2 (Spring): 161-74.
Ziemba, W. T., and Hausch, D. B. 1986. Betting at the Racetrack. New York: Norris M.
Strauss.
Ziemba, W. T., and Hausch, D. B. 1987. Dr. Z's Beat the Racetrack. New York:
Morrow.
Ziemba, W. T., and Vickson, R. G. (eds.). 1975. Stochastic Optimization Models in
Finance. New York: Academic Press.


Arbitrage Strategies For Cross-Track Betting On Major Horse Races

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Arbitrage Strategies For Cross-Track Betting On Major Horse Races

Hochgeladen von

Copyright:

Verfügbare Formate

Arbitrage Strategies for Cross-Track Betting on Major Horse Races

Author(s): Donald B. Hausch and William T. Ziemba

This content downloaded from 62.122.76.78 on Mon, 23 Jun 2014 07:20:17 AM