Documente Academic
Documente Profesional
Documente Cultură
In an efficient market news is incorporated into prices rapidly and completely. At-
tempts to test for this in financial markets have been undermined by the possibility
of information leakage unobserved by the econometrician. An alternative is to switch
to laboratory conditions, at the price of some artificiality. Potentially, sports betting
markets offer a superior way forward: traded assets have terminal values and news
can break remarkably cleanly, as when a goal is scored in soccer. We exploit this con-
text to test for efficiency, applying a novel identification strategy to high-frequency
data. On our evidence, prices swiftly and fully update.
1
We are grateful to several colleagues for their interest in and advice regarding this research, particularly Pete
Kyle and Kevin Sheppard. Versions of the paper were presented at the Oxford-Man Institute of Quantitative
Finance, Oxford University’s Postdoc and D.Phil. Workshop in Economics, the Conference in Honour of
David Hendry, Oxford, 2007, and the IMA’s First International Conference on Mathematical Modelling in
Sport, 2007. We received many helpful comments at these events. Financial assistance from the Economic
and Social Research Council and the Economics Department at Oxford is also gratefully acknowledged.
2
I. Introduction
Applying efficiency tests in the real world, most investigations have centered on conventional
financial markets.6 For instance, and regarding public information (the second form of
2
The efficient markets hypothesis is most commonly associated with Eugene Fama (1965, 1970, 1991). Its
early origins can be traced back to the work of Louis Bachelier, who in 1900 studied the dynamics of stock
price behavior.
3
For illustration, weak-form efficiency rules out the possibility that technical analysis techniques could be
used to produce excess returns, though analysis of fundamentals still might.
4
By implication, under semi-strong form efficiency not only technical analysis but also fundamental analysis
will be powerless to deliver abnormal returns. With regard to the speed and completeness of updating,
the definition of semi-strong form efficiency given in the text is the strictest interpretation. Less strict
formulations exist whereby it is sufficient for efficiency that it not be possible to trade upon the relevant
subset of information in such a way as to earn above-normal profits.
5
Strong form efficiency suggests that no one can earn excess returns, not even with privileged information.
6
Vaughan-Williams (2005) offers a comprehensive review of the academic literature which has investigated
information efficiency in financial markets.
3
test), researchers have scrutinized the response of share prices to such things as stock splits
(Fama et al. 1969), the release of company results (Ball and Brown 1968, Beaver 1968),
merger announcements (Asquith 1983), and announcements about economic variables such
as the money supply (Waud 1970, Chen et al. 1986). Somewhat problematically considering
the objective of such enquiries, it can be hard to tell when news really breaks in financial
markets—it is difficult to rule out information leakage not observed by the econometrician.7
Some investigators have preferred to analyze markets in the laboratory, where conditions
can be tightly controlled (Chamberlain 1948; and more recently Plott and Sunder 1988; List
2004). But while experimental settings can eliminate some concerns their artificiality raises
others: What trading experience do subjects have? Are they appropriately motivated?8
Potentially, sports betting markets offer a superior lens for efficiency studies, especially where
news arrival is the focus. Unlike laboratory experiments, these are real markets, with par-
ticipants that are properly motivated and often experienced. Contracts on sports outcomes
have well-defined terminal values (unlike equities and other financial securities) and this helps
simplify analysis.9 Moreover, and most importantly, major sports news often breaks remark-
ably cleanly. For instance, once a soccer game has kicked off the most significant innovation
in information concerns the scoring of a goal, and this event becomes common knowledge
at a single identifiable point in time. This is particularly so where a game is televised, as
many now are. Unfortunately, until very recently it was impractical to base efficiency stud-
ies around sports news; wagering was tightly controlled by traditional bookmakers (dealers),
who posted prices, updated these infrequently, and allowed betting only up until kick-off.
Only in 2000 did this picture change, but the change when it came was dramatic: harnessing
7
Some market participants may be party to (some part of) the content of announcements before these “go
public” (Jarrell and Poulson 1989). See Worrell et al. (1991) for an illustrative discussion of leakage in the
context of layoff announcements.
8
See Levitt and List (2006) for a recent consideration of factors affecting the generalizability of laboratory
findings, including the extent to and manner in which subjects are scrutinized in their decision-making.
9
As Gil and Levitt (2006) explain, in the context of sports betting there is no need to take account of
expectations regarding future events (e.g. dividend streams) when pricing the asset.
4
the Internet and inspired by conventional financial exchanges, betting exchanges began to
appear, offering online order-driven markets in bets. For the first time, prospective punters
could view a live order book, buy and sell bets at prevailing prices, submit limit orders, and
do all this “in-running” (as play unfolds). The development proved popular with customers,
and, in a few short years, the leading exchanges established themselves as serious operations.
The dominant player, Betfair, already sees trading comparable in intensity (if considerably
smaller in volume) to activity on the world’s leading financial exchanges.
Recognizing the research opportunity exchange betting presents, Gil and Levitt (2006) ana-
lyze data from the Intrade exchange (www.intrade.com), which operated markets for sports-
related bets until recently.10 Considering fifty matches from the 2002 Soccer World Cup, the
authors report that Intrade prices, though they respond strongly to a team scoring, trend
for ten to fifteen minutes after the goal is registered. On the face of it, this drift appears
reminiscent of the post-news drift found in some financial market studies. The finance litera-
ture describes post-event drift as “the tendency of individual stocks’ performances following
major corporate news events to persist for long periods in the same direction as the return
over a short window—usually one to three days—encompassing the news announcement it-
self” (Jackson 2006).11 Gil and Levitt interpret the drift they observe in Intrade’s markets
as evidence of informational inefficiency—prices, they suggest, update sluggishly to the news
of a goal. Unfortunately, though their analysis is insightful in many ways (for one, the data
come disaggregated at the individual trader level, and the authors are able to document
10
To avoid difficulties with US law, which considers wagering on sports to be gambling, Intrade-Tradesports,
a single operator at the time of the 2002 World Cup, has since split into two separately registered companies
with different activities: Tradesports deals with sports betting, whereas Intrade now operates as a prediction
market focused exclusively on non-sports events.
11
The literature concerning price drift after various corporate events is reviewed by Kothari and Warner
(1997), Fama (1998), and Daniel, Hirshleifer, and Subrahmanyam (1998). To mention just two studies,
Patell and Wolfson (1984) find that following dividend and earnings announcements it takes five to ten
minutes for profitable trading opportunities in individual equities to disappear, and Chan (2003) examines
returns to a subset of stocks after public news about them is released and finds evidence of post-news drift.
5
the endogenous emergence of market makers), as a test for semi-strong form efficiency it is
limited. There are two reasons for this.
The first relates to data quality: Intrade soccer markets attract very few traders (on average
just 75 per game); these people make very infrequent trades (an average game attracts 100–
200 trades and features several minutes in which trades do not occur); and betting volumes
typically are low (just $1.5m is traded in total across the full set of fifty matches).
The second limitation concerns methodological approach. Gil and Levitt interpret as ev-
idence against efficiency the fact that prices in these soccer markets drift for some time
following a goal. But, crucially, price drift could be evidence for efficiency in sports markets
that are “in play.” Consider that, once a soccer match (or other timed sports encounter) is
underway, participants update to major news, such as a goal, but also continually update
to the passage of time without a goal. Merely the ticking down of the clock gives rise to a
continual flow of information, and so prices should continually drift. Depending on the phase
of play considered, the current score, and probably many other factors, purely time-related
drift (evidence for efficiency) could be substantial, and properly identifying goal-related drift
(evidence of inefficiency) in its presence is likely to be a non-trivial affair. The authors ac-
knowledge (p.7, footnote 7) that playing time without a goal constitutes minor news, but,
from their analysis and discussion of results, they do not perceive this to present a serious
complication. We suggest, to the contrary, that the identification issue must be resolved, if
proper inferences are to be drawn regarding informational efficiency.
In this paper, we deploy a new data set to test for semi-strong form efficiency and propose a
simple identification strategy to deal with time-related drift. Our data were extracted second-
by-second, from the live order book of the Betfair exchange (www.betfair.com). Betfair
arrived onto the market in 2000, becoming one of the very first betting exchanges. It has
grown to be overwhelmingly the largest; its turnover of $50m per week accounts for ninety
per cent of all exchange-based betting activity worldwide and it currently has over a million
6
registered users. Two million trades a day—six times the number of trades on the London
Stock Exchange—are processed through Betfair markets, which now cover a vast variety of
events, mostly sporting.12 Our data consist of “in-running” prices and volumes related to
betting on the outcomes of English Premiership soccer matches played during the 2006–07
season. These markets are particularly heavily traded (on average $6m is traded per game).
Our identification strategy is notably simple but attractively clean. It involves exploiting
the (virtually) newsless window provided naturally by the fifteen minute half-time interval.
Since the clock stops during this break in play, time-related drift is not possible and the
matter of identifying potential goal-related drift is drastically simplified. Concretely, we
study matches with goals that arrive on the cusp of half-time—there are thirty nine in our
sample—and test the hypothesis that such news immediately shifts prices but does not cause
these to drift during the interval. We are unable to reject this hypothesis: prices update
so swiftly and completely that the news of a goal is fully digested by the time the break
commences. This is so even where the goal occurs just moments before the end of play. On
this evidence, we suggest that these markets are remarkably semi-strong form efficient.
We can think of two possible concerns with the approach we adopt, both related to the
potential for our efficiency finding to be an artefact of our half-time identification strategy.
First, one might suspect that our inability to detect sluggishness in updating is due primarily
to a lull in trading over the break—perhaps prices are stable because no one trades during
half-time. We dismiss this suggestion by tracking trading activity—for matches with a
goal just before half-time, betting interest, far from dissipating, actually ramps up over the
break.13 Second, and as a consequence, a vigilant reader might worry that particularly heavy
half-time trading is driving our result—perhaps these markets are efficient only during the
break. Based on our overall analysis, we strongly suspect that this is not the case, though
12
“Betfair Makes Online Odds on AC Milan, Hillary Clinton, Weather,” Bloomberg News, September 6, 2005.
Article available online at: http://quote.bloomberg.com/apps/news?pid=nifea&&sid=a8Y11XQeIcyY
13
With the exception of a match between Wigan and Liverpool, in which by half time Liverpool had built up
a virtually unassailable 4-0 lead. This early domination no doubt killed interest in the match odds market.
7
with the data we have it is difficult to counter this formally. Certainly, we observe that these
markets are well-traded during minutes of play, even when goals just before the break excite
half-time trading. Future work may deploy statistical modelling to construct counterfactual
time-related drift, and use this to take a look at updating “in-play.”14
The rest of the paper is organized as follows. The next section provides further background
on the betting industry and discusses Betfair in more detail, whilst Section III describes the
particular data set used in this study. Section IV discusses estimation strategy and presents
the main findings. Concluding remarks are set out in Section V. Appendix A contains
supplementary materials referred to in the main text.
Traditionally, betting markets have been run by a closed community of licensed dealers,
known as bookmakers. Bookmakers are similar to market makers in financial markets; they
establish and maintain liquid markets by quoting prices at which they will deal. In betting,
the prices are termed “odds” and the most common type of bet is known as a fixed-odds
bet. Suppose party A wishes to back (bet on) some outcome and party B wishes to lay (bet
against) the same. Then under a fixed-odds bet, A agrees to pay B a certain amount (the
backer’s “stake”) if the outcome fails to materialize, and B agrees to pay A the same stake
multiplied by pre-agreed (hence “fixed”) odds if instead it does. For example, A might stake
$100 at odds of 3 : 1 (‘three to one’) that Argentina will win the World Cup. In this case,
she collects $300 from B if Argentina succeed, but otherwise B keeps her $100 stake. When
betting with bookmakers, customers are restricted to backing outcomes only; the bookmaker
plays the role of party B, taking the lay side to every bet.
14
Croxson and Reade (2007) lay the groundwork for research in this direction by attempting to reverse
engineer goal arrival. The paper fits a bivariate poisson model to English Premiership data following the
approach laid out in Karlis and Ntzoufras (2003). Preliminary analysis suggests that probabilities implied
by Betfair prices closely track those suggested by the selected poisson process.
8
Odds relate inversely to the probabilities associated with particular outcomes.15 For instance,
odds of 3 : 1 imply a view that Argentina is three times more likely to fail than to succeed (a
25% probability of Argentine victory).16 Bookmakers rely on in-house gambling experts to
assess the likelihood of different outcomes and to compile a set of odds accordingly. As the
event draws closer the odds can be adjusted, reflecting the arrival of relevant information and
the bookmaker’s desire to maintain a balanced book. Odds are described as “fair” when the
implied probabilities sum to one, but built into the set of prices offered by the bookmaker
is a return for liquidity services (known in betting circles as the “overround” or “vigorish”)
such that the sum of probabilities exceeds one. In 1999, this bookmaking model was still the
only model of betting, and bookmakers belonged to an exclusive and profitable club. In the
UK, one of the world’s key betting markets, it was illegal for anyone other than a licensed
bookmaker to accept bets and a handful of major players (William Hill, Ladbrokes, Coral)
dominated the market. The overround stood at a healthy twenty two per cent.17
The arrival in 2000 of online betting exchanges marked a revolution in the industry. The
leading exchanges are essentially order-driven markets in fixed-odds bets, allowing individ-
ual punters to bet with each other directly, thereby disintermediating the bookmaker. This
means that exchange bettors can and do lay individual outcomes, contrary to the standard
bookmaking model. In addition, exchanges allow customers to place bets “in-running,” once
an event is underway. This is felt to have created a significantly more exciting betting expe-
rience. Typically, customers are charged a small commission for exchange betting services,
but the exchange does not otherwise impose any overround. Compared with bookmakers’
odds, exchange prices, at least for popular events, have tended to be highly competitive.18
15The interpretation of betting prices as probabilities is a somewhat debated area. The interested reader is
referred to Wolfers and Zitzewitz (2006) and the articles cited therein, particularly Manski (2004).
16In this example, the odds are quoted in so-called fractional form. An alternative is to quote decimal odds,
in which case the stake is included in the quoted multiple, so that 3 : 1 becomes 4. This is convenient because
the implied probability is then obtained simply by inverting the decimal odds, e.g., 1/4 = 0.25.
17
Merrill Lynch Research, 17 January 2006.
18
The dominant exchange, Betfair, claims its prices are on average 20% more generous than bookmakers’.
Ozgit (2005) confirms the competitiveness of Betfair’s pricing in the Basketball markets, although he does
9
The real hurdle for exchanges has been to achieve sufficient liquidity. Betfair was one of the
first exchanges to market, and is now by far the largest. It levies a standard commission of
5% on winning bets, falling to 2% for the heaviest users. Betfair’s early entry into the mar-
ket and its decision to run with a model much closer to a standard financial exchange than
some of its competitors (notably Flutter.com) are thought to have been pivotal its success.19
Volumes on the exchange are estimated to have doubled from $5.23bn to $11.06bn between
2003 and 2004, and almost doubled again between 2004 and 2005.20 These growth rates are
well ahead of those for the gambling market generally. Figure I benchmarks Betfair to the
world’s largest financial exchanges in terms of trade frequency.21 Betfair processes around
two million trades a day—six times the number of trades on the London Stock Exchange.
The selection of markets Betfair offers is vast and covers most sporting events of popular
interest, together with many non-sporting events (such as key political events and reality
TV). Horse racing dominates exchange turnover, followed by soccer. Within soccer betting,
customers can place bets related to the ‘Outright Winner’ of a particular league or tourna-
ment, or the ‘Top Scorer’ of the competition, for instance. Meanwhile, ‘Match Odds’ markets
allow betting on the outcome of individual games, by backing (betting on) or laying (betting
against) the ‘Home Win’, ‘Away Win’, or ‘Draw’. For those with less conventional betting
draw attention to the failure of exchange sometimes to offer sufficient liquidity at inside (best) prices. He
notes that punters wishing to place large bets can be better off taking their custom back to bookmakers
rather than “walking down the order book” at the exchange, accepting increasingly unattractive prices to
get large orders filled.
19
Flutter.com, founded in February 1999 by American management consultants, was the first person-to-
person betting site. Des Laffey (2005) analyzes some of the operational and marketing differences likely to
have led to Betfair’s dominance over (and eventual merger with) its main rival, despite the Flutter managing
to attract a comparatively huge amount of financial backing for its launch: “Flutter believed that they could
thrive by facilitating social bets between friends, for example about who would win a game of golf, and
also limited the value and frequency of bets allowed.” “Flutter’s website was not based around the Betfair
idea of matching pools of money from backers and layers, instead requiring a complete match between a
single backer and a single layer. Multiple transactions on an event by a punter on Flutter were also treated
separately which led to inefficiency whilst the Betfair model recognised mutually exclusive outcomes.”
20
Merrill Lynch Research, 17th January 2006.
21
This figure is based on analysis undertaken by Stephen Roman, Analyst, FXCM, New York, commissioned
and reported by prediction markets blog midas.org.
10
..
.... ...
.....
..... .. . . . ..
. ... . .....
. .........
.
.
.
.. .............
..........
. . . . . .. ..
.....
.. ........... .......
.... .....
.
.. .
.
..... ...............
.
..... .. . . . . . .
Betfair
.. .......... ....... ... ........ . ... ..
.. ... ...... ....... ......... ...... ......
... ... ......... ............. ..........
.. ............ ..
50 .. . . .. .
. . . .
. .. . . . ................
.
.
.
. . . .... ......
.
... ...
... ...
..
...
.
...................
.......
....
..
...
.. .. ICE/NBOT
.... .... .... .... .... .... .... .... .... .... .... ....
... .... .... .... . .... .... .... ....
.... .... .... .... .... .. .... ...... ...............
... .. . ... .. .. ......................... .... ............... ............... ... ........
..
.. ............... .... ..... ..............
. .. .... .... .... . .
. .. .. ............... ....... . ... . ........
........................... ........................... ....................... ....................... ............... .......................... ....................... ............... .. ....... ...............
LSE
0
02/2005 08/2005 02/2006 08/2006 02/2007 08/2007
preferences, there are markets such as ‘Over 2.5 goals,’ ‘Half-Time Score’ and exotic bets,
such as Asian Handicaps and Multipliers.
Suppose a user wished to bet on the outcome of a recent Premiership encounter between
Arsenal and Manchester United. Figure II shows the order book shortly before kick off.22 The
order book indicates, among other things, that should she wish to back Manchester United,
she might immediately stake up to $16,784 at odds of 3.3, and up to a further $53,140 at
slightly less attractive odds of 3.25. Betfair uses decimal odds which are inclusive of stake.
So a $10 bet to back Arsenal at odds of 2.58 would result in a gross return of $25.80 ($15.80
profit plus $10 stake). All odds are displayed from the backer’s point of view. Thus, 2.6
and $16,289 on the ‘Lay’ side of that market implies that someone (or some combination of
22
The standard view of the Betfair order book shows the best three prices (and corresponding available
volumes) on the back and the lay side. By clicking on the team name it is possible to view the full order
book showing any prices and volumes available beyond the first three steps of the book, along with historical
prices charts for each selection.
11
users) has submitted limit-orders hoping to back Arsenal asking for odds of 2.6 (i.e., slight
better than the prevailing market odds). If she were to accept $10 of this, by placing a lay
order at 2.6, the user would be betting against Arsenal and risking $26 to win $10.
III. Data
The data deployed in this paper comprise second-by-second prices and volumes from Betfair’s
‘Match Odds’ markets for 185 matches played in the English Premier League during the
2006–07 season.23 As discussed in the previous section, ‘Match Odds’ markets offer betting
on each of the possible outcomes of the event (in the case of soccer games, the Home Win,
Away Win and Draw). Our web crawler visits the order book related to bets available on
each of these outcomes and collects the following variables on a second-by-second basis:
(1) timestamp;
23
Due to technical and practical limitations it was not possible to collect betting data at high frequency for
all 380 Premiership matches that took place during the season. The representativeness of the final sample
is discussed later in this section, while Table 4 contains summary information on the matches sampled.
12
(2) the game outcome to which the order book relates (e.g. Home Win);
(3) the best three “back” prices and the volumes available to bet at each of these prices;
(4) the best three “lay” prices and the volumes available to bet each of these prices;
(5) whether the market is “in-play;”
(6) whether the market is “suspended;”
(7) the total cumulative volume traded on the Match Odds market for this game.
The market for a particular match is “in-play” when that match is in progress. As mentioned
in the previous section, Match Odds markets for English Premiership games are heavily
traded, particularly during the games themselves. Across the sample as a whole, an average
total of $6.23m is traded per match. On average, 49% of the total trading volume is bet
in-running, which equates to $22,390 traded per minute and $373 per second. Many English
Premiership games are now televised (either on Sky Sports or Premiership Plus subscription
channels) and the sample features an interesting mix of televised and untelevised encounters.
Television coverage tends to boost associated Betfair trading significantly, with non-televised
matches trading around $298 per second, and televised matches $766.24 Summary statistics
for the 185 matches in our sample are reported in Table 4 of the appendix.
Betfair briefly suspends its in-play soccer markets at kick-off and then briefly again upon the
occurrence of any “Material Event.”25 A Material Event in the soccer context is defined as a
goal being scored, a penalty being awarded or a player being sent off (the awarding of a red
card).26 Goals are the most important news event and the market reaction to goals forms the
focus of the paper. Goals arrive fairly infrequently; the average number of goals per match
in our sample is 3.06, implying 0.034 goals per minute of play. During a goal-related trading
suspension Betfair discards any unfilled orders, so clearing the entire order book. When the
24
The pay-per-view Premiership Plus matches, presumably because they have lower viewership, see a slightly
lower volume of matched trades per second ($739) than the standard subscription Sky Sports matches ($783).
25
The market suspension is on average 56 seconds long, with a median suspension of 53 seconds and standard
deviation of 20 seconds. 83% of stoppages lie within one standard deviation of the mean.
26
http://content.betfair.com/aboutus/?product=exchange®ion=GBR&locale=en GB&brand=betfair
13
...................................
...
..
...
1.00 .
...
.
...
..........................................................
...........................
...................... .
.. .. ... ... ...........................................
... ... .... .... ..........................................................................................................................................................................
.. ... .. .........
... ... ... ...
.... ...
0.75 ... .. ...
... ... .... ....... ....
2 ...
...
...
... ..
...
.... ..
..................
... .... ... ... .......
...
...
................
....... ... .... .... ........
.. .. ....
...
...
........ ...
...... ... . 0.50 ... ... .......
... ..... ............................................................................................................................................................... ... ... ......
... ...... ...... .....................................................................................
......................................................... ...... ...................... ........
1 ...
..
...
...... ......
.... ....
...... ......
...
...
......... ...... .....
...
...
...... .....
.... ....
... ..... .....
....
...
....... .......
...... ......
...
...
0.25 ...
... ...... ......
...... ......
... ...... ...... ... ..
... ..... ..... ... ... ...... ......
... .
..... ..... . ... ... ...... .....
... .
..... ..... . .. ... ...... ......
.. ..... ..... ............................... .. ..... .....
0 0.00 .. ..
17:17 17:32 17:47 18:02 18:17 18:32 18:47 19:02 17:17 17:32 17:47 18:02 18:17 18:32 18:47 19:02
order book reopens the odds have shifted, reflecting updating by the market about the teams’
relative chances of winning and the likelihood of a tie. Figure III illustrates the suspension
of trading and subsequent price updating in a recent Premiership encounter between Bolton
and Tottenham Hotspur.
The left hand panel plots the “in-running” time series for the best price to back a Bolton
win, and this price is converted into an implied probability in the right hand panel.27 Bolton
scores two goals in quick succession, at 17:25 and 17:29, and the suspension of trading and
removal of all unfilled orders following each goal results in vertical lines at these points in
time. In each case, once the market is reopened the book quickly fills up with new orders at
new prices, and this updating manifests in clear jumps in price and implied probability.
Data concerning the exact timing of Material Events, such as goals and cards, were collected
from SportingLife.com. Sometimes matches kick off slightly later than officially planned
(15:03 rather than 15:00, for example) and typically a few minutes of extra time (usually
called ‘injury time’) are played at the end of each half to allow for minutes lost to injuries
and other stoppages. Unfortunately, it is not possible to obtain official match timings and
these were instead estimated by allowing on average one minute of first-half injury time and
27Implied probabilities are computed as 1/(decimal odds).
14
three minutes of second-half injury time.28 Where the data suggest something extraordinary
has happened (a serious injury, or some kind of other disturbance that held up play), this
is accounted for by internet-based research to determine the nature of the incident and the
length of delay it created.29
Figure IV. Scatter plot of team appearance in the sample against average
volume traded on its matches. This provides a check on sample representa-
tiveness. Lack of positive correlation is ideal, suggesting that teams that are
more heavily traded tend not to be over-sampled.
.... .... .... .... .... .... ....
.. .. .. .. .. .. ..
... ... ... ... ... ... ...
..........................................................................................................................................................................................................................................................................................................................................................
140% .....
....
.....
....
.....
....
.....
....
.....
....
.....
....
.....
....
... ... ... ... ... ... ...
.... .... .... .... .... .... ....
... ... ... ... ... ... ...
.... .... .... .... .... .... ....
... ... ... ... ... ... ...
... ... ... ... ... ... ...
...
Middlesbrough ... ... ... ... ... ...
... ... ... ... ... ... ...
...
...
• ...
...
...
...
...
...
...
...
...
...
...
...
... ... ... ... ... ... ..
...........................................................................................................................................................................................................................................................................................................................................................
120% ...
...
...
... Newcastle ...
...
...
...
...
...
...
...
...
...
... ... ... ... ... ... ...
...
...
...
...
• ...
...
...
...
...
...
...
... Man Utd
...
...
.... .... .... .... .... .... ....
Appearance ....
...
Sheff Utd
....
...
....
...
....
...
....
...
Arsenal
• ....
... • • ....
...
.... .... .... .... .... .... Chelsea ....
of
team x
...
...
...
• Fulham,Wigan
...
...
... ............
...
.
.
....
...
.
.
....
...
.
.
....
...
.
.
....
...
...
...
Aston Villa ... Watford Man City ........ .. . . . ..
(relative
100% • •
..........................................................................................................................................................................................................................................................................................................................................................................
... ... ..
.
....... ....
.... ...
.
.
.
.
. .
.
.
.
.
.
.
.
. ...
to ... ... . .
..
...
.
...........................
.
.
. .
.
. .
.
. ...
mean • ...
Bolton
...
...
•• ••• West Ham ...
...
...
.
..
.
....
.
..
.
........
. ..
.
..
....
.
.
..
• ....
.
.
..
....
.
.
..
...
...
...
team Reading ... .
.. Liverpool
.
..
.
..
.
..
appearance)
...
...
...
• • ....... ..
..........
.......
•
Blackburn Tottenham
... ... .
..
.
.
.
.
.
..
.
.
.
.
.
..
.
.
.
.
.
..
.
.
.
.
...
...
...
Charlton ... ...
.
.
. .
.
. .
.
. .
.
. ...
... Everton .... .... .... .... .... ...
.... .... .... .... .... .... ....
.... .... .... .... .... .... ....
80%
...
.
.
...
...
.
.
...
•
.....................................................................................................................................................................................................................................................................................................................................................
...
.
.
...
...
.
.
...
...
.
.
...
...
.
.
...
...
...
... Portsmouth ... ... ... ... ... ...
... ... ... ... ... ... ...
... ... ... ... ... ... ...
... ... ... ... ... ... ...
... ... ... ... ... ... ...
... ... ... ... ... ... ...
... ... ... ... ... ... ...
... ... ... ... ... ... ...
... ... ... ... ... ... ...
... ... ... ... ... ... ...
... ... ... ... ... ... ...
... ... ... ... ... ... ...
.. .. .. .. .. .. .
. .
..................................................................................................................................................................................................................................................................................................................................................... . . . .
60% ...
...
...
...
...
...
...
...
...
...
...
...
...
...
.... .... .... .... .... .... ....
. . . . . . .
number of goals scored, in which case the fact that games in our sample feature a higher-
than-average number of goals could be concerning. However, it turns out on inspection that
the correlation between goals in a match and trading interest is a minuscule 0.01. These
things considered, we argue that our sample is representative and provides a good basis for
investigating market efficiency.
If Betfair markets are semi-strong form efficient, prices (and the probabilities these imply)
should rapidly and fully update to major news that very obviously is publicly available. In
16
this section, we assess the immediacy and completeness of the Betfair price response to goals
scored during the soccer games in our sample.
The top left cell of Table I indicates that a goal on average induces an immediate twenty
four point increase in the scoring team’s probability of winning. This statistic is in line with
previous findings on goal impact. For instance, in Gil and Levitt (2006) a World Cup 2002
goal induces a change of between twenty and thirty points in the implied probability that
the scoring team wins. Intuitively, the other cells in Table I suggest considerable variation
behind this simple average. Rows (2)–(7) consider subsets of goals according to scorer type
(whether the scoring side is playing at home or away, for instance) and lateness in the game,
and columns (2)–(7) subdivide goals according to the goal difference they create. Goals
that change the status quo outcome (e.g. from a draw to a win) have the greatest impact;
goals that result in the scoring team leading have a greater impact (column 3) than goals
that level the scores (column 2); these goals on average immediately add respectively thirty
six and fifteen percentage points to the probability that the scoring side wins. Goals that
merely boost a side’s lead tend to have smaller effects. For example, a goal that extends a
team’s lead to two goals (column 5) increase its win probability by just thirteen points.32
The impact is greatest where outcome-changing goals occur towards the end of the game,
30Recall from the previous discussion that Betfair suspends the market briefly in the event of a goal.
31In computing this shift, we look at the market price immediately before the goal is registered and compare
this to the price shortly after the goal-induced trading suspension, once reasonable liquidity has returned to
market.
32Interestingly, Gil and Levitt (2006) report that World Cup goals which bring teams level are associated
with shifts nearer twenty per cent, whereas goals that put a team one or two goals ahead induce shifts of
around thirty percentage points. It is not inconceivable that the nature of domestic and international soccer
can account for much of the discrepancy between our respective findings on goal impact; McHale and Scarf
17
as would also be expected. A goal scored very late on in the game—after the eighty ninth
minute—and which puts a team in the lead adds a dramatic eighty eight points to the
probability that they go on to win the game (row 7, column 3).
Whereas we can be comfortable on the strength of this evidence that there is an immediate
reaction to goals in these markets, it is somewhat more complicated to ascertain whether
the jumps observed reflect complete Bayesian updating or simply mark the beginning of an
updating process that takes some time to complete. To pursue this question of completeness,
we might think about comparing the new post-jump price level with the level several minutes
later (assuming no further goals in that time), and construing any significant difference
between these as evidence of informational inefficiency. For instance, Gil and Levitt (2006)
find that Intrade soccer betting prices drift for ten to fifteen minutes after goals scored in
the 2002 World Cup and interpret this as suggestive of sluggish updating. However, to
identify inefficiency in sports betting markets that are “in-play” one is forced to confront an
interesting and non-trivial complication: some amount of price drifting is perfectly consistent
with, and indeed evidence for, market efficiency in this setting as rational participants would
be expected continually to Bayesian update to the minor news inherent in the passage of
(2006) document important differences between these two types of soccer game, notably that there is less
disparity in the quality of the competing teams in domestic football games.
18
To appreciate the nature and inevitability of time-related drift in in-play prices, consider that
as match time elapses the likelihood of further goals fades, and the probability thus grows
that the current standing of the teams will come to reflect the final match outcome. In the
absence of further goals, the probability associated with the status quo should drift upwards
over time, eventually reaching unity by the end of play, whereas probabilities associated with
the other possible outcomes of the game should drift downwards towards zero. Figure IV
illustrates time-related drift using the prices associated with two Premiership fixtures.
The left panel relates to an encounter between Newcastle and Chelsea. Depicted is the in-
running probability (as implied by the best available back price on Betfair) that Chelsea,
strong favorites at the start of the match, ultimately will take the match. The match ended
in a goalless draw and note that the win probability clearly drifts downwards as the clock
ticks down and the chances recede of Chelsea breaking the deadlock. The right panel relates
to another match involving Chelsea, in this case a game at home to Tottenham Hotspur a few
19
Time Number
of of
goal goals
40th min 4
41st min 6
42nd min 4
43rd min 2
44th min 7
45th min+ 16
weeks earlier. As the first half slips away without a goal, the probability that Chelsea wins
again begins to drift downwards. At 13:55, just after the second half commences, Chelsea
scores to move into the lead. As expected, the probability of a Chelsea win jumps upwards in
response to the goal but continues then to drift upwards over the remainder of the game. This
upward drift, at least partially, will reflect rational updating to the closing window of time. It
may also reflect some sluggishness in updating to the goal, and therein lies the identification
challenge. In cases such as this, we might consider modelling rational time-related price
movement as a way to identify possible drift associated with inefficiency. Modelling time-
related drift is not trivial, however; it will depend on various factors, including the current
scoreline (e.g. the magnitude of any lead) and the stage in the game. In this paper, we
introduce a simpler and arguably cleaner identification strategy; we exploit the (virtually)
newsless window provided naturally by the half-time interval. Concretely, we propose to
look at games where goals are registered on the cusp of half-time (henceforth “cusp goals”)
and see whether over the break in play, where time-related drift cannot be present, we detect
any news-related drift. Our data set contains thirty nine such matches.
Figure VI takes a closer look at the distribution of “cusp goals” by time of arrival.33 Sixteen
of the goals occur in the final minute of the second half (including any stoppage time), which
33Six matches feature two cusp goals. We count only the time of the second goal in these cases.
20
Implied probability of Manchester United win. US$ volume traded each minute.
1.00 ...
... ←Half→ ...
...
. ..............................
................. ........
... Time ... ... ......
.... ... ... .... .....
200,000
.. .. .. .. ...... ←Half→
.......................................................................................................... ......
...... Time
0.75 ... ...
... ...
... ......
... ......
.....
......
... ... ... ...... ......
... ... ... ...... .....
. ..
. . ....
. ......
................................................................................................................................................. .... ....
. .... ........ ......
...... .. ....................... . .. .. ..... ......
0.50 ......
............................ .. ..
..... ..
. .
.. ......
. ......
......
......
...... ...
...... ...
.. ......
.
.... .......
. ......
......
100,000
...... ...... ... ... ...... ......
...... ...... .. . ....
. ......
...... ..... ... .. ......
. ......
0.25 ......
......
...... ...
...... ...
.. ......
.
.. ......
.
......
......
...... ...... ... . .....
.
. ......
.... .. .
...... ..... .. .... ........ ......
...... ...... ... ... ...... ......
...... ........ ... ... ........ .......
0.00 0
16:01 16:16 16:31 16:46 17:01 17:16 16:01 16:16 16:31 16:46 17:01 17:16
is encouraging, since the closer the goal to half-time, the stronger the test for semi-strong
form efficiency. A further seven goals arrive during the penultimate minute of play.
A visual look at the price series for a few of these matches is suggestive of efficient updating.
Consider, first, Figure VII, in which Tottenham play at home to Manchester United.
This match kicks off just after 16:00 and the plot in the left hand panel shows the probability
of a Manchester win, as implied by the best Betfair back price. This probability is 56% at
the beginning of the game (Manchester being favorites to win) but begins to drift downwards
as the first half progresses without a goal. By the 44th minute the probability has fallen
to under 50%. Then right at the end of the first half Manchester scores to take the lead
and the market is suspended briefly. When it reopens moments later, the probability has
jumped up to 77%. Almost immediately the whistle blows for half-time. Over the fifteen-
minute interval that follows, the implied probability appears to remain remarkably constant
at this 77% level, suggesting that updating to the goal was immediate and complete. A
legitimate concern might be that such evidence for efficiency is an artefact of our half-time
identification strategy: perhaps prices appear not to continue to update over half-time only
because trading interest drops off during the break. In the right hand panel we report trading
21
1.00 ←Half→
.
.....
.
..... ←Half→
... ... 60,000 Time
Time ..
....
...
...
... ...
........... ... ...
0.75 .
.
. .
. .
. .
. .
..
...
. . .
......................................................................................................................................... .
.
.
.
.
.
...
...
..... . ................................ . . . .
.
..... . ..
...
..
..
..
..
..
..
..
....
..
..
...
. ............................... . ...
.....
.....
. ....
. ..
....................... ..
... ..
.
.
. ...
...
40,000
...... ... .. ...
0.50 ......
.....
... ...
... ... ...
...
...... ... ... ...
...... ... ... ...
...... ... .......
... ................................................................................................................ ...
......
..... .
... ... ..
.
.
.
.
.
.
.
... ....... 20,000
0.25 .....
......
. .. .. ..
... .. ..
. ...
...
..... ... ... ... ...
...... ... .. .. ...
..... .. ... ... ...
...... ... .. ..
.... ......... .... ....
0.00 .
0
15:03 15:18 15:33 15:48 16:03 15:03 15:18 15:33 15:48 16:03
activity. Certainly for this game, the market is actively traded throughout half-time. In fact
trading interest appears to step up during the interval in play.
For contrast, Figure VIII illustrates the case of an upset. Here, the ex-ante favorite concedes
a goal just before half-time. The post-goal probability (again, as implied by the best Bet-
fair back price) seems somewhat more volatile but still there appears no obvious trending.
Meanwhile, trading during the half-time break is again heavy. Looking across all matches
in our sample in which a goal occurs just before half-time, betting interest always ramps
up during the interval; on average $386 is traded per second of play but $422 per second is
traded over the half-time break.34
The apparent lack of half-time trending in such diagrams, despite heavy trading during the
break, constitutes prima facie evidence that the market is semi-strong form efficient. The
rest of this section implements regression analysis to test this formally. Pursuing a two-step
approach, we begin by determining whether there is in fact any half-time drift, by using OLS
to estimate the following regression:
34Inmatches without a goal on the cusp of half time $387 per second is traded on average during the match
compared $247 during half-time.
22
Here, t denotes seconds during half-time; i and p index particular match (e.g. Arsenal vs.
Manchester United) and market (e.g. home win).
The second step is to discern whether, to the extent that there is price drift during half-time,
this drift is connected to the arrival of the goal. For this, we estimate the following regression
model:
(2) α
b1,i,p = β0 + β1 goali + β2 2goalsi + εi,p ,
where goali takes 1 if goal takes place between the 40th minute and the half-time whistle,
zero otherwise, and 2goalsi takes 1 if two goals take place in this interval, otherwise zero.36
Additionally, because it is reasonable to suspect that pooling goals scored closest to the
interval with goals scored many minutes before the break (to which the market has had
several minutes to adjust by the time the break begins) would somewhat mask the effects of
later goals, impulse variables for each minute from the 40th to the 45th are created (denoted
g40i , . . . , g45i ) and the following regression implemented:
35Although some of these minuscule trends are significant, there is no relationship between significance and
cusp goals; the tiny correlations between goal occurrence and trend significance (there is more than one
correlation as there is the home win, draw and away win price series, and the maximum such correlation is
-.091) are just as often negative as positive.
36There are six matches where two goals are scored immediately before half-time, highlighted in bold typeface
in Table A.2.
23
Table II. Testing the effect of cusp goals on the estimated half-time price
trend coefficient. The number of goals in the five minute period prior to
half time is reported first, followed by the minute in which the cusp goal was
scored. The t-statistic is reported, with the probability of a Type I Error
given in parentheses below.
(3) α
b1,i,p = β0 + β1 g40i + β2 g41i + β3 g42i + β4 g43i + β5 g44i + β6 g45i + εi,p .
PT
These two regression models are also run on the residual variance from (1), σ 2
bi,p = T −1 t=1 εi,p,t
in order to ascertain whether pre-half-time goals impact market volatility. Table II displays
the t- and F- statistics (with the probability of a type I error in parentheses) relating to
the trend regressions (2) and (3), while Table III reports the same output for the variance
regressions.
From Table II all the impulse variables in regression models (2) and (3) are insignificant;
goals in the five minutes before half-time have no impact on the drift observed during the
half-time interval. Goals in the 44th minute are close to exerting a significant effect, but
jointly all such goal variables are highly insignificant from the F-test statistic. No other
individual variable for any type of goal is close to significant.
From Table III it appears that cusp goals can have an impact on the volatility of the price
of a home team win; although the effect of one cusp goal is insignificant, two such goals
has a considerable impact, with a t-statistic of 2.45.37 Considering just individual goals,
37As the F-test, to a crude approximation, is the square of the t-test, the joint significance displayed in the
significant F-test statistic can mainly be attributed to the two goals effect.
24
Table III. Testing the effect of cusp goals on the estimated variance of half-
time prices. The number of goals in the five minute period prior to half time
is reported first, followed by the minute in which the goal was scored. The
t-statistic is reported, where a value above 2 denotes statistical significance.
44th minute goals are individually significant, with a t-statistic of 2.12, although the joint
F-statistic of all five impulse variables remains insignificant. It appears, however, that the
volatility impact of cusp goals is limited to the home-win market, as no other variables are
significant in Table III.
We do not interpret these selected significant coefficients for volatility as evidence of market
inefficiency for a number of reasons. First, many statistical tests are being carried out, and
when taking a 5% significance level, one expects spurious rejections around 5% of the time;
of forty eight t-tests, two fail, which is under 5%. Further, the effect is a variance effect
and not a trend effect, and so not evidence for post-news trending. Also, the volatility effect
appears to be strong only where two cusp goals occur in a single game, an uncommon event.
As such we take the broad sweep of the evidence presented in Tables II and Table III as
providing support for informational efficiency: these soccer markets rapidly and completely
impound the news of a goal.
V. Concluding Remarks
The recent emergence of online betting exchanges has made it possible to obtain high-
frequency data relating to bets placed “in-running” (during a live sports event). This implies
a fertile new setting for empirical work, and in particular, it paves the way for a cleaner look
25
at the topic of market efficiency. A market that is semi-strong form efficient updates swiftly
and fully to publicly available information. A problem for those seeking to put this to
the test in financial markets has been the possibility of news leakage not observed by the
econometrician. In sports, however, major news (such as a goal in soccer) tends to break
comparatively cleanly. We exploit this characteristic of sports events to offer a fresh study
of efficiency. Prices for soccer-related markets are extracted from the live order book of the
largest online betting exchange, Betfair.com, and tested for efficiency in relation to the arrival
of goals. A complication particular to this exercise relates to the hypothetical difficulty in
determining whether any price drift following a goal should be interpreted as sluggishness in
updating (and hence evidence of inefficiency), or be considered simply an efficient response to
the passage of time without a goal (goalless periods of play being themselves price-relevant
news). To overcome this identification issue, we exploit the naturally newsless half-time
interval—we study matches where goals arrive on the cusp of the half-time break. Our
findings suggest that prices rapidly and completely impound news.
References
[1] Asquith, P., “Merger Bids, Uncertainty and Stock Holder Returns,” Journal of Financial Economics,
XI (1983), 51–83.
[2] Bachelier, L., Theorie de la Speculation, (Paris: Gautier-Villars, 1900), reprinted in English in The
Random Character of Stock Market Prices, P. Cootner, ed. (Cambridge, MA: MIT Press, 1964).
[3] Ball, R. and P. Brown, “An Empirical Evaluation of Accounting Income Numbers.” Journal of Account-
ing Research, VI (1968), 159–78.
[4] Beaver, W., “The Information Content of Annual Earnings Announcements,” Journal of Accounting
Research, VI (1968), 67–72.
[5] Chamberlain, E., “An Experimental Imperfect Market,” Journal of Political Economy, LVI (1948),
95–108.
[6] Chan, W.S., “Stock Price Reaction to New and No-News: Drift and Reversal after Headlines,” Journal
of Financial Economics, LXX (2003), 223–260.
[7] Croxson, K. and J.J. Reade, “Using a Bivariate Poisson Model of Goal Arrival to Explain Exchange
Betting Prices,” Research Paper, Economics Department, University of Oxford (2007).
[8] Chen, N., R. Roll, and S. Ross, “Economic Forces and the Stock Market,” Journal of Business, LVIIII
(1986), 383–403.
[9] Daniel, K.D., D. Hirshleifer, and A. Subrahmanyam, “Investor psychology and security market under-
and over-reactions,” Journal of Finance LIII (1998), 1839–1885.
[10] Gil, R. and S. Levitt, “Testing the Efficiency of Markets in the 2002 World Cup,” Research Paper,
University of Chicago (2006).
26
[11] Fama, E.F., “The Behaviour of Stock Market Prices,” The Journal of Business, XXXVIII (1965), 34–
105.
[12] —–, L. Fisher, M. Jensen and R. Roll, “The Adjustment of Stock Prices to New Information,” Inter-
national Econonzic Review, X (1969), 1–21.
[13] —–, “Efficient Capital Markets: A Review of Theory and Empirical Work,” The Journal of Finance,
XXV (1970), 383–417.
[14] —–, “Market efficiency, long-term returns and behavioral finance,” Journal of Financial Economics
XLIX (1998), 283–306.
[15] Garicano, L., Palacios-Huerta, I., and C. Prendergast,“Favoritism Under Social Pressure”, Review of
Economics and Statistics, LXXXVII (2005), 208–216.
[16] Jarrell, G. and A. Poulson, “Stock Trading Before the Announcement of Tender Offers: Insider Trading
or Market Anticipation?,” Journal of Law, Economics, and Organization, V (1989), 225–248.
[17] Jackson, A., “Unifying Underreaction Anomalies” Journal of Business, LXXIX (2006), 75–114.
[18] Karlis, D. and I. Ntzoufras: “Analysis of Sports Data By Using Bivariate Poisson Models,” The Statis-
tician, XXV (2003), 381–393.
[19] Kothari, S.P. and J.B. Warner, “Measuring long-horizon security price performance,” Journal of Finan-
cial Economics XLIII (1997), 301–339.
[20] Laffey, D., “Against the Odds—Case Study of Betfair.com,” The International Journal of Entrepreneur-
ship Education, III (2005), 99–114.
[21] Levitt, S. and J. List, “What Do Laboratory Experiments Tell Us About the Real World?,” Journal of
Economic Perspectives, forthcoming, 2007.
[22] List, J., “Testing Neoclassical Competitive Theory in Multilateral Decentralized Markets,” Journal of
Political Economy, CXII (2004), 1131–1156.
[23] Manski, C., “Interpreting the Predictions of Prediction Markets,” NBER Working Paper No. 10359,
2004.
[24] McHale, I. and P. Scarf, “Forecasting international soccer match results using bivariate discrete dis-
tributions,” Management and Management Sciences Research Institute Working paper, University of
Salford, 2006.
[25] Ozgit, A., “Auction versus Dealer Markets in Online Betting,” Working Paper, Department of Econom-
ics, UCLA, November 22, 2005.
[26] Patell, J. and M. Wolfson, “The Intraday Speed of Adjustment of Stock Prices to Earnings and Dividend
Announcements,” Journal of Financial Economics, XIII (1984), 223–252.
[27] Plott, C. and S. Sunder, “Rational Expectations and the Aggregation of Diverse Information in Labo-
ratory Security Markets,” Econometrica, LVI (1988), 1085–1118.
Harry V. Roberts. ”Stock Market ’Patterns’ and Financial Analysis: Methodological Suggestions.”
Journal of Finance, 14 (March, 1959), 1–10.
[28] Roberts, H.V., “Statistical versus clinical prediction of the stock market,” Working paper, University of
Chicago, 1967.
[29] Vaughan-Williams, L., Information Efficiency in Financial and Betting Markets, (Cambridge: Cam-
bridge University Press, 2005).
[30] Waud, R. N., “Public Interpretation of Discount Rate Changes: Evidence on the ’Announcement Effect.”
Econometrica, XXXVIII (1970), 231–250.
[31] Wolfers, J. and E. Zitzewitz, “Interpreting Prediction Market Prices as Probabilities,” CEPR Discussion
Paper No. 5676, (2006).
[32] Worrell, D., W. Davidson III, and V. Sharma, “Layoff Announcements and Stockholder Wealth,” Acad-
emy of Management Journal, XXXVIII (1991), 662–678.
27
Goals Traded
Date Home Away Time TV H A Upset $m In-play H-T?
8/19/06 Arsenal Aston Villa 15:00 N 1 1 D 4.84 49% Y
8/19/06 Bolton Tottenham 17:15 PP 2 0 N 4.49 64% Y
8/19/06 Sheff Utd Liverpool 12:45 SS 1 1 D 9.97 48% Y
8/20/06 Chelsea Man City 16:00 SS 3 0 N 7.95 29% Y
8/20/06 Man Utd Fulham 13:30 PP 5 1 n/a 3.98 23% N
8/22/06 Tottenham Sheff Utd 20:00 N 2 0 n/a 2.82 24% Y
8/22/06 Watford West Ham 19:45 N 1 1 D 1.38 35% Y
8/23/06 Charlton Man Utd 20:00 SS 0 3 N 6.75 45% Y
8/23/06 Middlesbrough Chelsea 20:00 N 2 1 n/a 3.82 42% Y
8/26/06 Charlton Bolton 15:00 N 2 0 n/a 0.93 27% Y
8/26/06 Fulham Sheff Utd 15:00 N 1 0 n/a 0.80 26% Y
8/26/06 Liverpool West Ham 12:45 PP 2 1 N 7.06 69% Y
8/26/06 Man City Arsenal 17:15 PP 1 0 Y 10.33 57% Y
8/26/06 Tottenham Everton 15:00 N 0 2 Y 1.33 49% Y
8/26/06 Watford Man Utd 15:00 N 1 2 n/a 6.69 45% Y
8/27/06 Aston Villa Newcastle 14:00 N 2 0 N 3.15 60% Y
8/27/06 Blackburn Chelsea 16:00 SS 0 2 N 11.54 47% Y
8/28/06 Middlesbrough Portsmouth 20:00 SS 0 4 Y 8.42 43% Y
9/16/06 Bolton Middlesbrough 15:00 N 0 0 D 2.26 34% Y
9/16/06 Charlton Portsmouth 12:45 PP 0 1 Y 4.54 66% Y
9/16/06 Everton Wigan 15:00 N 2 2 n/a 3.77 37% Y
9/16/06 Sheff Utd Reading 15:00 N 1 2 N 1.01 55% N
9/16/06 Watford Aston Villa 17:15 PP 0 0 n/a 6.29 58% Y
9/17/06 Blackburn Man City 15:00 N 4 2 N 1.06 19% Y
9/17/06 Chelsea Liverpool 13:30 SS 1 0 N 11.63 55% Y
9/17/06 Man Utd Arsenal 16:00 SS 0 1 Y 14.38 43% N
9/17/06 Tottenham Fulham 15:00 N 0 0 D 2.01 27% Y
9/17/06 West Ham Newcastle 15:00 N 0 2 n/a 1.18 35% Y
9/20/06 Liverpool Newcastle 20:00 N 2 0 N 4.26 10% N
9/23/06 Arsenal Sheff Utd 15:00 N 3 0 N 5.79 26% Y
9/23/06 Aston Villa Charlton 15:00 N 2 0 N 1.02 29% Y
9/23/06 Fulham Chelsea 15:00 N 0 2 N 4.26 53% Y
9/23/06 Man City West Ham 15:00 N 2 0 N 1.02 24% Y
continued on next page. . .
28
Table A.2. Summary Statistics for Games with Cusp Goals. Games with
two goals are in bold typeface. In the TV column, N denotes an untelevised
match, SS represents a match televised on Sky Sports, and PP a match tele-
vised on Premiership Plus. In the final column the bracketed figure is the
volume traded per second during half time.
Date Home Away Time TV Goal Times $ Traded/sec(H-T)
8/26/06 Liverpool West Ham 12:45 PP L 41,44; W 11 773(879)
8/26/06 Man City Arsenal 17:15 PP M 40 936(1287)
9/17/06 Blackburn Man City 15:00 N B 18,44,68,89; M 39,45 32 (25)
9/17/06 Chelsea Liverpool 13:30 SS C 42 1013(1209)
9/24/06 Newcastle Everton 16:00 SS N 14; E 41 783(570)
9/30/06 Chelsea Aston Villa 15:00 N C 3; A 45 509(269)
9/30/06 Everton Man City 15:00 N E 44; M 90 61 (65)
10/14/06 Reading Chelsea 17:15 PP C 45. 1133(1229)
10/22/06 Tottenham West Ham 15:00 N T 45. 120(125)
11/5/06 Aston Villa Blackburn 14:00 N Ast 41,50 104(117)
11/11/06 Everton Aston Villa 15:00 N A 42 124(202)
11/11/06 Wigan Charlton 15:00 N W 13,41,78; C 52,90 50(53)
11/12/06 Arsenal Liverpool 16:00 SS A 41,56,80 1024(1583)
11/25/06 Aston Villa Middlesbrough 15:00 N A 45; M 43 128(66)
11/25/06 Bolton Arsenal 17:15 PP B 9,45,76; A 45. 1077(809)
11/26/06 Tottenham Wigan 15:00 N T 43,44,90; W 25 171(267)
12/2/06 Wigan Liverpool 15:00 N L 9,26,40,45. 160(4)
12/27/06 Charlton Fulham 20:00 SS C 19,42; F 13,90 853(749)
12/30/06 Bolton Portsmouth 15:00 N B 30,40,62; P 2,89 87(49)
1/13/07 Charlton Middlesbrough 15:00 N C 27; M 45,63,68. 58 (31)
1/13/07 Watford Liverpool 12:45 PP L 34,40,48 438(501)
1/20/07 Man City Blackburn 17:15 PP B 44,62,90. 720(592)
1/20/07 Middlesbrough Bolton 15:00 N M 6,10,23,43,84; B 25 113(182)
1/20/07 Newcastle West Ham 15:00 N N 45,53; W 18,22. 185(157)
1/20/07 Reading Sheff Utd 15:00 N R 44,50,70, S 77. 49 (95)
2/4/07 Tottenham Man Utd 16:00 SS M 45,48,54,77. 662(1222)
2/10/07 Chelsea Middlesbrough 15:00 N C 45,66,83. 240(252)
2/21/07 Everton Tottenham 20:00 N E 42; F 35,89 130(48)
2/21/07 Watford Wigan 19:45 N W 24; W 40. 56(66)
2/24/07 Charlton West Ham 15:00 N C 24,34,41,80 139(417)
3/31/07 West Ham Middlesbrough 15:00 N W 2,45. 95 (224)
4/1/07 Tottenham Reading 16:00 SS T 41 364(423)
4/7/07 Arsenal West Ham 15:00 N W 45 432(497)
4/7/07 Wigan Bolton 15:00 N W 32; B 44,68,73. 32 (9)
4/9/07 Watford Portsmouth 12:45 SS W 28,45,51,73; P 16,81 568(356)
4/14/07 Arsenal Bolton 15:00 N A 31,45 ; B 11 94 (26)
4/14/07 Middlesbrough Aston Villa 15:00 N M 13; A 45,70,77 20 (17)
4/17/07 Arsenal Man City 19:45 N A 12,73,80; M 41. 460(342)
5/6/07 Arsenal Chelsea 16:00 SS A 43; C 70 1061(1448)