Sei sulla pagina 1di 16

POST POSITION BIAS:

An Econometric Analysis
of the
1987 Season at Exhibition Park

Sandra Betton'
Department of Finance, U n i v e r s i t y of B r i t i s h Columbia
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.
Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

Abstract
This paper investigates the eldstence of a post position bias during the 1957 racing season at
E-xhibition Park. The econometric techniques of Probit and Pooled Cross-sectional Time Series
analysis were used to analyze the relationship between the odds ranking, post position, and
final position of a horse. The length of the race does not significantly improve the fit of any of
the models. The results of this analysis indicate that the post position significantly adds to the
information reflected in the odds rankings. The significance of the post position bias tends to
increase as the number of horses taking part in the race increases.

Ever since the time of Augustus Caesar and the Roman chariot races, the betting public has
been searching for a system to maximize their profits or more realistically, to minimize the losses
incurred at the track and in so doing, have observed certain features inherent in horse racing. These
features include, believed biases due t o the length and narrowness of the track, the track conditions,
race distance and the commonly held belief that the post position exerts a significant bias.
This paper examines the relationship between finishing position, odds ranking and post position.
The odds rankings were found to more fully reflect the post position bias the fewer the horses
involved in the race. The relationship between post position bias and abnormal profits were not
examined in this paper. (See Canfield, Fauman, and Ziemba [1987] for a detailed examination of
this issue).
The importance of a solid understanding of the relevant biases arising a t particular race tracks
is underscored by the amounts wagered and paid out at these race tracks. O n average, a bettor at
E.xhibition Park loses between 15 and 20 percent of the amount wagered due to track commissions
and taxes (Table 1). Table 2 indicates that certain uninformed betting rules may allow one to do
better than average, ie, t o lose less than average. Table (7) summarizes the dollars paidback for
uninformed betting categorized by the number of turns in a race. Therefore, it is in the bettor's
best interest to understand and. where necessary, price the results of any observed bias.

1 Causes of Post Postion Bias


The expected post position bias arises from the physical design of the modem race track. Most
Borth American horje racing tracks are oval or elliptical and have banked turns. The oval nature of
the track forces an outside horse t o run further than an inside horses. Thus, if four horses started
'This paper originated b9 a term paper in Dr. Ziemba's course on Speculative Investments. Without implicating
them. I would Like to thank the edirors for their help in prepanng this paper for publication.

511
512 S. BETTON

Vancouver, British Columbia


1979 1986 I
Days ol racing 108
Totd attendance 996,573 875,336 PER PERSON BASIS
1979 1986
Totd Wagered $107.00 $146.00
Track commission and
breakage $10,529,474 $13,611,058 Co&ion + Bredrage 10.56 15.55
Pmvincid tax $7,4714,696 $8,965,941 Tax 7.50 10.24
Paybsck $88.94 8120.21
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.
Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

from post positions one through four and ran at the same pace, the extra 10 lengths run by the #3
horse in a three turn race would preclude him from victory.2
The second cause of a post position bias is the banking of the track. Banking can be advan-
tageous as it promotes drainage, thereby, helping to prevent deep water from accumulating on the
track. This provision can, however, hinder the inside horses as water would tend t o accumulate
next to the inside rail with the result th at the outside horses are running on a relatively dry track
while the inside horse's track is still wet and slow.
Banking, however, can also be a major disadvantage in dry conditions t o any horse forced to run
up that gradient. If only the turns are banked then the outside horse effectively gallops up and then
down a hdl, significantly affecting the power available from the animal. For example, a gradient
of 1:lOO causes a 10% reduction in a horse's pulljng ability (see Marks 11941)). Unfortunately,
muscular energy used running up a hill is not regained running back down the hill.
One would, therefore, expect the post position bias to be most significant on short tracks with
less straights to enable one t o jockey for position on the inside rail. This is particularly so on short
tracks with very sharp turns.

2 The Data Set

Data were collected from the results of racing a t Exhibition Park as reported in the Vancouver
Sun and Province newspapers for the period &om April 13, 1987 to October 18, 1987. Ninety-seven
races were excluded from the sample for the following reasons:
1. two horses were reported to have the same post positions.

2. cases where normal betting would be affected due to coupled entries, late scratches and in
one case, where no "show" wagering was permitted.
3. one day of racing was eliminated as only four races were run before all racing was suspended
due to poor conditions.
'Canfield. Fauman, and Ziemba (19871 show that for a semicircular turn, the inside horse travels TT feet around
the turn. A horse m horses outside must travel an additional r m w feet where w is the average width of a horse.
POST POSITION BIAS 513

Exhibition Park
April 13. 1987 - October 18, 1987

Betting on Poet Position I Betting on Oddr Flank

Post # of Odds #of


Pmn Bets To Win To Place To Show Rank Bets To Win ToPlacc To Show
1 1062 78.03 86.59 78.09 1 1089 85.47 93.92 97.15
2 1062 86.76 85.62 81.05 2 1068 88.70 85.68 87.44
3 1082 77.00 78.19 81.98 3 1042 79.82 83.32 87.91
4 1062 98.48 8336 90.52 4 1062 80.66 76.69 85.93
5 1061 87.81 82.66 80.27 5 1054 78.82 79.22 78.79
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.

6 1027 81.70 82.86 79.20 6 1037 82.83 78.23 78.30


Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

7 935 72.70 73.50 70.73 7 935 86.27 85.79 78.22


8 790 73.66 M1.77 76.68 8 787 67.05 76.59 63.42
9 641 64.80 64.21 63.57 9 641 107.25 70.28 57.18
10 484 74.68 70.18 73.78 10 483 35.53 36.46 38.39

4. cross-track betting on races which actually took place a t other tracks.

One thousand and sixty-two races, representing 9239 horses, comprised the final sample. Sev-
enteen of these races resulted in dead heats.
This paper does not examine the post position bias in conjunction with the track condition bias.
The period in question was unusually dry for Vancouver and this is reflected in the distribution of
track conditions. A 95% confidence interval places the average track condition between fast and
good.3
Final win odds were ranked4 as the actual odds quoted are functions of the relative sizes of the
win pools and therefore, may not be comparable between different races and days.
Exhibition Park is an interesting track for examining post position bias as i t is short, only 5/8
of a mile long. Canfield, Fauman, and Ziemba (1987) examine the issue of post position bias in
relation to the length of the track and the races involved.
A second characteristic of racing a t Exhibition Park that is useful in examining the post position
bias is the high proportion of large numbers of horses.in races. Over 70% of the races involved
more than 7 horses (see Figure 1).

3 Analysis
3.1 Graphical Analysis
A graphical analysis of the results of horse races by post position was conducted t o determine
if a post position bias seemed likely. Figure 2 compares actual and expected results of the first
10 post positions. T he expected number of finishes by each post position was determined under
the assumption that, in the absence of a post position bias, each post position is equally Likely to
place first, second, or third. The In Top Three class refers to the percentage difference between
expected and actual number of times the post position finished in the top three places.
See Canfield, Fauman and Ziemba (19871 for a detailed analysis of the off-track effect on the post position bias.
* o d d s ranking 1 = race favourite, n = nth favourite
514 S. BETTON

Figure 2 indicates that the inner four to five post positions appear, on average, t o do better
than the outer four post positions. Th e significance of this apparent bias was examined using a
t-test to compare the average observed post position of the top three horses t o the average post
position expected under the hypothesis that no post position bias exists and therefore. each post
position is equally likely to finish in any p ~ s i t i o n . ~

Table 3: ODDS AND POST POSITION OF T H E T O P TH R EE HORSES


Post Position of Top Three H o n a I Odds Ranldng of Top T b Horn-
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.
Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

Observed Mean 5.3634 4.7871 4.8327 4.8193 4.3158 4.0815 4.1822 4.1152
T-statistic .8541 4458 4345 -2.2427 -1.6022 -10.6609 -5.5218 -12.0657
N 38 761 269 1068 38 761 269 1068

The results in Table 3 indicate that the post position bias is significant for first and second
place; but, not for third place. Table 3 also summarizes similar d at a for odds ranking.
The post position bias is far less dramatic than the odds bias (see Figure 3 and Table 3).
These preliminary results support the findings that a t this track the win odds are useful tools in
the prediction of horse race results.
If the odds were determined independently of the post position, one would expect that, on
average, the odds rankings would then be equally distributed across post positions for a given size
of race. The uniform distribution expectation arises due to the random manner of post position
assignment.
The equal distribution expectation was examined for races having 10 horses. This size of race was
selected as it comprises the largest group of races. Figure 4 indicates that the uniform distribution
hypothesis is not supported by the data. Th e hypothesis of no post position effect, ie. that the
odds rankings are uniformly distributed across post positions, was tested using the Chi-square test
of goodness of fit. T he results summarized in Table 4 indicate that the post position bias affects
the odds ranking, and has a major impact on inside horses in these races. Th e hypothesis of a
uniform overall distribution can be rejected a t the 0.05 level for all field sizes.

In an N horse race, each post position has a 1 f N chance of placing in any position. T h e expected post position
+
of the win, place, or show horse, in that particular race, is,therefore, (N I ) / ? .
POST POSITION BIAS 515

ble 4: T E ST O F UNIFORM 0 )S RANK1NG DISTRIBUTIC I


xz Test of Gt nesi of Fit
O b s w e d x2
10 horses i n the race T horses in t h e race Observed xz
p a t position 1 17.01 post position 1 33.09
p a t position 5 29.18 post position 4 24.95
post position 10 53.05 post position 7 26.05

Critical values: Critical values:


a = .05 x2 = 16.91 a = .05 xa = 12.59
Overdl observed xz = 187.18 Overall observed xz = 190.78
Critical x2(99,.05) = 123.22 Critical xz(48..05) = 65.17
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.

The finding that the odds ranking distribution of the outside post position, compared to the
Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

inside position, in a ten horse race is more significantly different &om uniform lends support t o the
notion of the odds taking into account the post position of the horse. A possible rationale for these
observations is that bettors determine the probability of a horse winning before considering the post
position and then adjust odds rankings for the post position drawn. Typically, the participants in
a horse race are announced several days prior to the actual race. In contrast, the post position
of the horses is usually announced on the day of the race. In this manner, the bettors are able
to evaluate the horses probability of winning before adjusting for the post position drawn. This
hypothesis would be tested by examining the change in odds rankings due to the announcement of
post positions. Unfortunately, this d ata is not available.
The issue of whether, or not, th e odds account for the post position of the horse was further
analyzed by examining the differences between the odds ranking of different post positions within
a race.
The basic hypothesis is that if post positions matter, one would expect the odds rankings of
horses to be more similar the closer their post positions were. The null hypothesis being tested is
that the true absolute average difference in odds rankings is less than or equal t o one. As we are
starting from the pole position horse (post position #1) and moving outwards, we would expect
the difference to be negative. Essentially, we test the joint hypothesis t h at odds rankings increase
as we move outwards from the pole position and that the true difference between the odds rankings
of two adjacent post positions is less than or equal t o one.
The results of the paired t-tests are given in Table 5 Th e underlying assumption of normality
of the differences is not significantly violated as indicated by the normal probability plots of these
differences.

In Table 5 , the mean difference between odds rankings in post position A and post position
B (in the same races) are found a t the intersection of row A and column B. T h e value labeled
( p ) refers to the probability of observing th at difference under the null hypothesis that the true
difference was less than or equal t o minus one. For example the mean difference between the odds
rankings of post position 1 and 2 is -.70 while the probability of observing that value under the
null hypothesis is 1.
The null hypothesis of a difference of less than one was used in preference t o the more standard
null hypothesis of a difference of zero due to the odds ranking procedure. Using the zero null in
a situation where the odds are ranked would lead to a bias toward rejection as any infinitesimal
difference in the odds would result in odds rankings which differ by a t least one, possibly even
516 S. BETTON

Table 5: TEST O F ODDS RANKING DIFFERENCES BETWEEN POST POSITIONS IN T H E


SAME RACE
POST 2 3 4 5 6 7 8 9 10

1 -.7 -1.0 -1.5 -2.0 -2.2 -2 6 -3.1 -3.5 -3.7


(p) 1 .5 0 0 0 0 0 0 0
2 -.3 -.8 -1.4 -1.5 -1.9 -2.3 -2.6 -2.8
(;I 1 .95 0 0 0 0 0 0
-.5 -1.0 -1.2 -1.5 -2.0 -2.3 -2.5
(P) 1 .4 .04 0 0 0 0
-.5 -.7 -1.0 -1.4 -1.7 -1.9
(P) 1 99 0 0 0 0
5 -.2 -.5 -.a -1.2 -1.4
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.

(PI 1 1 -89 -17 .04


Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

6 -.4 -.a -1.1 -1.3


(P) 1 .95 22 .05
7 -.4 -.7 -.9
(PI I .Y5 .76
8 -.3 -.5
(P) 1 1
9 -0.0

more. Unless the odds are identical, the observed difference between odds rankings is at least one
in absolute value. T he minimum absolute difference in odds rankings which can be observed in
practice when the underlying odds are not identical is one.
The Table 5 shows that a significant difference did not exist between the odds rankings of
horses in the last five post positions. Adjacent post positions, results on the diagonal of Table
5 did not have significantly different odds rankings. As the distance between horses increases,
moving toward the upper right of the table, significant differences between the odds rankings begin
to appear.
One can, therefore, conclude that a post position bias appears to exist and that the odds
determination process appears to consider th e post position of the horse. Th e question of whether
or not the odds rankings adequately reflect the post position bias remains unanswered.

3.2 Statistical Analysis


3.2.1 Individual Analysis

Typical econometric techniques cannot be used to estimate a dichotomous dependent variable,


the horse either finishes in the top three or it does not, as one typically finds estimated probabilities
greater than one. Probit and logit functions provide a technique for ensuring that the estimated
probabilities lie between 0 and 1. Logit analysis is based on the logistic function while probit
analysis is based on the cumulative normal probability function.
A probit analysis was undertaken to examine the impact of odds rankings, post positions and
number of horses on the probability of a horse finishing in the top three positions. The number of
horses was included in the analysis as the graphical evidence indicates that the strength of the post
position bias is related to the number of horses in the race. As one is dealing with odds rankings
the number of horses in the race would also be required to give a comparative value to the odds
ranks.
POST POSITION BIAS 517

Table 6: W U L T S OF PROBIT A N W S I S
Total number of observations 9239
Number of hones in one of the top three positions 3194

Estimation One: Odds and number of hones


% comct prcdictioo. = 69.5%

CocK. Asymptotic t-ratio


Od& rank -.I8123 -31.723
Numb. of horse -.05174 -5.382
Constant ,91482 10.715

Log-likelihood function = -5320.2,LRT = 1276.21 with 2 degrccs of freedom


by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.
Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

Estimation two: Odds, post position and number of horses


96 ofpredictioru correct: 69.4%
COeR. Asymptotic t-ratio
Post Position -.01135 -2.063
Odds rank 48018 -31.415
Number of horsw -.04677 -4.720
Constant .92133 10.782

Log-likelihood function = -5318.1.LRT = 1277.47 with 3 d e p a of ireedorn


Estimation three: Odds, post pmition, length of rare, and number of horse
% of predictionr correct: 69.3%

COCK Asymptotic t-ratio


Port Position -.01135 -2.063
Od& rank -.18018 -31.414
Length of race .W170 0.159
Number of hones -.04677 -4.720
constant .%I959 18.054

Log-Likelihood function = -5318.1,


LRT = 1277.49with 4 degrees of keedom

Likelihood Ratio Tat (LRT) of HO; no true relationship exists

Note: An uninformed bettor wagering that any horse would


f i s h in the top three would be correct on average 35% of the time.

Probit analysis was selected instead of logistic analysis as the probit analysis assumes that t h e
probability distribution is approximately normal. This assumption was not unreasonable due t o the
large number of observations. The analysis estimates the cumulative probability density function:

F ( X ' P ) = G(T 5 X ' p )


where G is the standard normal distribution. Inference is carried out asymptotically in a probit
model as the maximum likelihood estimator is only asymptotically normally distributed. Th e
results of this analysis appear in Table 6 .
Estimation two, which includes the post position, is significantly better, in terms of fit, than
estimation one, which does not include the post position. As Shazam (the computer code available
a t the University of British Columbia [White, 19781) uses Maximum Likelihood Estimation, hy-
potheses can be tested using a likelihood ratio test. The test comparing estimations one and two,
at the .01 significance level, had a critical value of 6.3 while the observed value was 4.2. Similar
results were obtained using the odds rather than the rankings.
518 S. BETTON

A bettor, randomly selecting horses, would correctly identify one of the top three horses, on
average, 35% of the time. The percentage of correct horse selections rises t o nearly 70% when the
odds rankings and post positions are taken into account.
The length of the race, which reflects the number of turns involved in a race and therefore, the
extra distance an outside horse must run, was included in the analysis in an attempt t o identify the
cause of the post position bias. The Iength of the race did not significantly add to the explanatory
power of any of the models estimated. It would, therefore, appear that the information contained
in the length of the race has been included in the odds rankings. The post position bias appears
to be caused by something other than the extra distance run by outside horses.
It would appear that the odds do partially adjust to account for the post positions of the horses
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.

involved, as illustrated by the insignificance of the length of the race. However, there is still a
Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

significant amount of useful information available from the post position.


The results, thus far, indicate th at both the strength of the post position bias and the reaction
of the odds ranking, are related in some way to the number of horses in the race.
The question of whether the apparent failure of the odds rankings t o fully account for the
post position bias is due to an inefficiency in the market, or is due t o a combination of many
different probability generating processes, remains unanswered. But, it seems reasonable that races
of different sizes might have different win generating processes.

3.2.2 Serial R e l a t i o n s h i p
A pooled cross section time series analysis was undertaken to examine the finish position as a
function of odds rankings and post positions on the basis of race size. This analysis was also
undertaken to evaluate the commonly held belief th at certain post positions are hot during the
course of a day of racing.
In this type of analysis, each race is taken to be an event, while the cross sectional units are
the post positions.
The model estimated is cross-sectionally hetero- scedastic and time wise autoregressive. This
analysis did not examine the probability of payoff, but, rather used the finishing positions as the
dependent variables. This pooled time series analysis was conducted on each size of race in order
to examine the relative importance of post position in different race sizes.
Pooled cross-section time-series estimates the following model:

Y; = xip + fi (1)

With cross sectional units represented by i , with i = 1,.. . N and time series denoted by

= (x,l,...x , T )
x; = ( X i * ,,...Xi,=)

The finishing position of the horse in race t starting from post postion X;,t is denoted by Y,,t. For
i = 1, ... N

; = ( ~ i . 1 ,... i,T)

i,t = Pici,t-I + e,,t


POST POSITION BIAS 519

Other conditions include:

E(ei,t) = 0
E(&,tej,t) = 0
E(ei,tej,,), = 0 for 1 # s
Assumptions underlying the model are:

1. The coefficients are the same for each post position.


by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.
Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

2. The errors follow a first order autoregressive process. In the estimated model, the autoregres-
sive process is assumed to be the same for each post position.
3. The variance of E can vary across different post positions.

4. The disturbances are contemporaneously correlated. (Judge et al, 1985).

The results of this estimation are summarized in Table b .Similar results were obtained using
odds rather than the odds rankings.
Including the post position significantly improves model fit for races having 7, 9 and 10 horses.
In all estimations, the autoregressive parameter, p , was not significantly different from zero. This
finding does not support the con- tention that certain numbers become hot during the day;
rather, that the results would lend support t o the hypothesis that the results of previous races do
not affect the outcome of the current race.
The general lack of serial behaviour of the post position bias is, however, consistent with the
fact that horses of sufficiently variable abilities are randomly assigned t o their post positions.
If all the horses are of equal ability, and there is no post position bias, we would expect, on
average, that each post position would win 1/N times where N is the number of horses in the race.
If the post position bias is significant, we would expect that certain positions would consistently
perform better and tha t these results would be correlated over different races. If, however, the horses
are not of equal ability, it is possible for a post position bias to not display a significant correlation
between races while still having a significant impact on the win probability of the different horses.

4 Summary and Conclusions


1. Th e results of the probit analysis indicate that the post position significantly adds t o the
information available from the odds rankings in determining the probability of a horse placing
in the top three positions.
2. We would expect, from the results of the probit and graphical analyses, that the post position
would be significant for some field sizes, but, not necessarily all. The pooled cross section
time series results support this contention. It is shown that if there are more than six horses
in the race, the post position becomes an important factor in addition t o the win odds.

The results of this analysis must be used with caution due to the very small data set and the
lack of pricing analysis. See Canfield, Fauman and Ziemba [1987] for a discussion of profit making
520 S. BETTON

a t the race track. The primary purpose of this paper was descriptive, essentially t o determine if
the post position bias exists and whether the odds rankings appear to have compensated for that
bias. Similar results were obtained when the analysis was repeated using the odds rather than the
odds ranking.
One must bear in mind that any statements generalized from this paper must take into account
that the data were only from the 1987 racing season a t Exhibition Park.
This paper has shown that knowledge of the post position significantly improves the information
available from the odds rankings. The relatively low overall explanatory power of these models
suggests that more is unknown than known in the determination of racing results.
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.

Table 7: DOLLARS PAIDBACK PER $100 WAGERED USING UNINFORMED BETTING


Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

Exhibition Park
April 13, 1987 - October 18. 1987

Races categorized by number of t w .

TWO TURNS THREE TURNS

POST #of # of # of
POS'N BETS WIN PLC SHW BETS WIN PLC SHW BETS WIN PLC SHW

1 40 116.5 93.00 68.38 755 79.30 84.65 79.02 267 68.58 91.12 76.93
2 40 107.88 85.88 67.13 755 86.21 83.15 81.47 267 85.15 92.55 81.95
3 40 44.25 86.25 76.13 755 74.01 80.09 81.61 267 90.36 7l.61 83.69
4 40 55.88 91.38 103.88 755 102.07 84.34 88.00 267 94.72 79.36 95.64
5 40 81.13 62.88 86.50 755 85.15 83.03 79.59 266 96.35 84.57 81.26
6 39 33.97 36.92 50.77 728 79.90 81.06 73.40 260 93.90 94.81 85.73
7 37 55.00 46.49 57.57 668 78.46 72.16 70.02 2-30 58.80 81.76 74.89
8 31 23.71 55.16 81.13 569 65.89 68.07 76.85 190 105.08 73.08 75.42
9 26 19.81 31.92 71.73 468 73.35 70.94 65.76 147 45.51 48.50 55.17
10 21 15.95 44.05 53.57 356 70.80 72.61 77.91 107 99.11 67.20 64.02

TWO TURNS THREE TURNS

# of # of
SHW BETS WIN PLC SHW BETS WIN PLC SHW

118.50 774 87.19 93.80 98.06 275 75.45 92.44 91.49


93.78 756 91.68 85.30 88.64 271 81.18 87.14 83.14
64.23 747 76.07 85.92 91.42 256 93.24 73.89 81.29
48.54 754 81.55 75.63 80.40 267 88.13 83.86 107.30
92.26 748 74.12 79.93 77.96 267 85.45 74.01 78.73
62.32 733 87.78 81.42 79.05 263 n.43 74.41 78.69
58.71 660 79.72 88.46 79.85 229 118.32 88.69 76.81
74.68 570 67.35 79.54 65.22 186 77.31 72.34 56.02
43.65 468 112.60 61.42 54.31 147 93.74 103.57 68.74
-43.33 354 32.87 26.95 --
36.03 108 51.16 65.37 -45.14
POST POSITION BIAS 52 1

5 References
Arnemiya, Takeshi. Advanced Econometrics. Harvard University Press, Cambridge, Massacusetts.
1985.
Canfield, Brian R., Bruce Fauman, and William T. Ziemba. "Efficient Market Adjustments of Odds
Prices to Reflect Track Biases". Management Science. Vol 33 #11. November 1987.
Judge, George G, W. E. Griffiths, R. Carter Hill,Helmut Lutkepohl, and Tsoung-Chao Lee. The
Theory and Pmctice of Econometrics. 2nd Edition. John Wiley and Sons. New York. 1985.
Krnenta, Jan. Elements of Econometrics. Macmillan Company, New York. 1971.
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.
Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

Marks, Lionel S. editor. Mechanical Engineer's Handbook. McGraw-Hill Book Company, New York.
p 1132. 1941.
White, Kenneth J. "A General Computer Program for Econometric Methods - SHAZAM". Econo-
metrica. January 1978. PP. 239- 240.
522 S . BE'ITON

Table 8: POOLED AND CROSS SECTIONAL TIME SERIES


Cmabscctions restricted to have the ssme s u t o m p s i i v e parameter.

N u m b e r of hona: 5 Number of hones: 6 Number of horses: 7


Number of racer = 34 Number of race = 92 Number of races = 145

Coeffic. T-ratio (166) Coeffic. T-ratio (548) Coeffic. T-ratio (1011)


Post pmition .Om9 1.21 .0307 .78 ,2941 3.18
Odds ,4111 5.79 2827 9.08 ,3449 11.67
Length ,0169 0.16 -.0023 -.05 .W04 .Ol
Constant 1.4082 1.82 2.1493 5.27 2.2496 5.72
Constant p: -.00268 Constant p: -.I324426 Constant p: ,011613
Log-Likelihood: -281.23 Log-Likelihood: -1037.80 Log-Likclihood:-2069.79
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.
Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

Coeffic. T-ratio (167) Coeffic. T-ratio (549) Coeffic. T-ratio (1012)


Post position .OW 1.23 .a31 .79 .095 3.19
Odds ,4109 5a 9 ,363 9.13 ,345 11.72
Constant 1.5227 4.72 2.133 10.47 2.253 13.21
Constant p: -.00268 Constant p: -.04425 Constant p: ,011613
Log-Likelihood: -281.24 Log-Likelihood -1037.80 Log-Likelihood:-2069.79

Cod%=. T-ratio (168) Cocffic. T-ratio (550) Coeffic. T-ratio (1013)


Odds .401 5.84 ,364 9.22 ,351 11.91
Constant 1.799 7.77 2.237 14.61 2.595 19.67
Constant p: -.wns Constant p: -.cQ359 Constant p:.OZ085
Log-Likelihood -282.100 Log-Likelihood-1038.11 Log-Likelihood: -2074.7i

Number of honer: 8 Number of honer: 9 Number of horse: 10


Number of race8 = 149 Number of races = 157 Number of races = 439

Coeffic. T-ratio (1188) Coeffic. T-ratio (1409) Coeffic. T-ratio (4386)


Post Postion -.0031 41 ,0447 1.78 .0521 3.71
Odds .3870 14.48 ,3730 14.96 .3933 28.15
Length .0034 .06 -.a003 -035 .W23 .07
Constant 2.7576 6.91 2.9159 7.77 3.0340 12.62
Constant P: ,05436 Constant p:.017206 Constant p: -.W1102
Log-Likelihood:-2577.90 Log-Lilrelihmd:-3236.79 Log-Likelihood: -10471.8

Coeffic. T-ratio (1189) Cocffic. T-ratio (1410) Coeffic. T-ratio(4387)


Post Position -.W3 -.11 .@I5 1.79 ,052 3.71
Odds 3 7 14.53 .373 15.01 ,393 28.18
constant 2.781 15.32 2.914 15.73 3.049 28.23
Constant p: .05436 Constant p: .01n1 COMtant p: -.W1101
Log-Likelihood:-2577.90 Log-Lik&hood:-3236.79 Lag-Likelihood: -10471.80

Coeffic. T-ratio (1190) Coeffic. T-ratio (1411) Coeffic. T-ratio (4388)


Odds ,387 14.59 ,374 15.09 .4w 28.92
Conrant 2.768 20.46 3.132 22.38 3.294 38.46
Constant p: ,05426 Constant p:.0195 Constant p: .rn2018
Log-Likelihood: -2577.91 Log-L&&ood:-3238.40 Log-Likelihood: -10478.70
Figure 1.
DISTRIBUTION OF RACE SIZES
BY NUMBER OF HORSES
9 (14.78%)
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.
Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

8 (14.03%)

10 (41.34%)

7 (13.65%)

>11 (4.24%)
(6 (3.30%)

DISTRIBUTION OF RACE LENGTHS


IN FURLONGS
6.5 (59.32%)

< 4 (3.77%)
>9 (3.58%)
Figure 2.
PERCENTAGE DIFFERENCE (POST BASIS)
ACTUAL AND EXPECTED d OF FINISHES
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.
Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com

1,
15

10

5
..,..
..:.:.
:.:..
i..:.
...::..
::.:.
.:::.
0
VI
P
N
-5

-25 I I I I I I I I I I I I

1 2 3 4 5 6 7 8 9 10
POST POSITION

FIRST PLACE SECOND PLACE 0 THIRD PLACE IN TOP THREE PLACES


Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.

525

.,:..,.,.,
.q: ,.:... .....
:.:. :......................
::.....:.:.::.::::..:< .,.,.,.:
. .., ..,,........,: ,I'.:.:,.::l
.:. .,,., .. .. ~

I
1

+I
Efficiency of Racetrack Betting Markets Downloaded from www.worldscientific.com
by NANYANG TECHNOLOGICAL UNIVERSITY on 05/11/16. For personal use only.

526

I
I
I
I

Potrebbero piacerti anche