Evaluating strange forecasts: The curious case of football match scorelines
| Published date | 01 May 2021 |
| Author | J. James Reade,Carl Singleton,Alasdair Brown |
| Date | 01 May 2021 |
| DOI | http://doi.org/10.1111/sjpe.12264 |
Scott J Polit Econ . 2021;68:261–285. wileyonlinelibrary.com/journal/sjpe
|
261
© 2020 Scottis h Economic Societ y
1 | INTRODUCTION
Forecasts fo rm a central part of everyday lif e; they are statements regar ding the probability of par ticular states
of nature occurri ng. In general, economi c agents have preferences ove r different states of nat ure, which can have
real consequences in money or other terms. As such, the evaluation of forecasts is important and in principle
ought to relate to agent s' preferences (e.g. Gr anger & Pesaran, 20 00). But for many variab les and contexts, t he in-
ability to obse rve or understa nd these preference s, plus the difficu lty of construct ing agents' loss func tions based
on what actuall y occurs, and allied with th e generally (quasi-)continuous n ature of macroeconomic varia bles, has
led to more stat istical measures of fore cast evaluation in most c ircumstances (Fawcett et al ., 2015).
In this study, we evalua te the forecasts of associatio n football (soccer) match outcomes. U ltimately, after all
the punditr y is said and done, there are t wo important aspec ts to the outcome of a football m atch: the result and
the scoreline. The result is a win for either team, or a draw (tie). The scoreline gives the exact number of goals
scored by each team . A football scoreline is thus a p air of non-negative integers, wh ich are correlated due to the
Accepted: 31 Augu st 2020
DOI: 10 .1111/sjpe.1 2264
ORIGINAL ARTICLE
Evaluating strange forecasts: The curious case of
football match scorelines
J. James Reade1| Carl Singleton1| Alasdair Brown2
1Departme nt of Economics, Uni versity of
Reading, Reading, UK
2School of Econo mics, Universit y of East
Anglia, Norwich, UK
Correspondence
J. James Reade , Department of Eco nomics,
University of Reading, Whiteknights
Campus, RG6 6UA , UK.
Email: j.j.reade@reading.ac.uk
Funding information
Economic and Social Research Council,
Grant/Award Number: ES/J500136/1
Abstract
This study analyses point forecasts of exact scoreline out-
comes for football matches in the English Premier League.
These forecasts were made for distinct competitions and
originally judged differently. We compare these with im-
plied probability forecasts using bookmaker odds and a
crowd of tipsters, as well as point and probability forecasts
generated from a statistical model. From evaluating these
sources and types of forecast, using various methods, we
argue that regression encompassing is the most appropri-
ate way to compare point and pro bability forecasts , and find
that both these t ypes of forecasts for football match s core-
lines generally ad d information to one another.
KEYWORDS
forecasting, prediction markets, regression models, statistical
modelling
262
|
READE EtAl.
common conditi ons faced by both teams in a match and t he fact that teams and their t actics will respond wi thin
matches to the goals s cored (or not) by their oppo nents (e.g. Heuer & Rub ner, 2012). The different st ates of nature
dictated by foot ball match outcomes mat ter significantly ; teams may progress in comp etitions, their fan s may gain
bragging rig hts, and bettor s may make returns (or losses). Wh ile the result gener ally determines the s tate of nature
(e.g. winning a round-robin or knock-out competition), the scoreline is sometimes the first tiebreaker after the
result. League positions and championships, where the teams are tied on cumulative point totals from results, are
usually determined by some function of scorelines (e.g. the difference between goals scored and conceded or
head-to-head reco rds between teams over multip le matches). Some cup competiti ons (e.g. the UEFA Champions
League) have scoreline-related tiebreaker rules, such as “away goals.”1 Even more fundamentally, the result is a
function of the scoreline.
The majority o f attention in the aca demic literature on fo recasting footba ll match outcomes has focus ed on the
result rather than the scoreline (Angelini & De Angelis, 2019; Forrest et al., 2005; Forrest & Simmons, 2000;
Goddard, 20 05). But scorelines also matte r. Many forecas ts are made regarding them, b oth formal and informal.
Based on our observations and estimation from the world's largest sports betting exchange in 2019, Betfair
Exchange, the exact scor eline in a football match is a popu lar outcome to predict and b et on: focusing on the state
of markets at the be ginning of important matche s (i.e. high liquidity markets w ith £1 million or more of matched
bets, e.g. th e English Premier League or c ompetitive internation als), for every £1.00 of bets m atched on the result
outcome of the match , approximately £0.20 is mat ched on the exact scoreli ne markets. This compare s with £0.70
bet on the total nu mber of goals scored in a match, £ 0.25 on the Asian Handi cap markets and £0.2 0 on the margin
(goal differ ence) between the two tea ms at the end of a match. Further, the se other mentioned outcom es of foot-
ball matches and p opular predictio n markets are all func tions of the final score line. As there are onl y three possible
outcomes for the r esult, and many times mor e potential outcomes for the s coreline, it follows that for ecasting the
scoreline is more difficult. Historically, the most likely result outcome from a football matc h is a home win (occurring
roughly 48% of the t ime), while the most likel y scoreline outcome is a 1–1 draw (occurring roughly 11% of the t ime).2
The scoreline is a strange variable . A general definition of strangeness is diff icult and may in fact vary by con -
text. But we thi nk that if one was written down , then it would contain parall els to some of the following aspec ts
of a football match sco reline forecast:
Non-standa rd: it is non-continuous, is m ade up of two non-negative i ntegers and generates a ra nge of import-
ant suboutcomes (e .g. the result).
Residual outcom e: the tie is a third outcome betwe en either team winning. De spite 1–1 being the most com-
mon outcome, it is a re sidual outcome.
Uncertaint y: a large number of potential event outcomes ensure that the most like ly has only about a 10%–
15% likelihood of occurring.
Fragility: the median number of goals is three, with a variance near to three, and over 10% of all goals are
scored in the fin al five minutes of matches.
Salience: the sc oreline determines the r esult of a football match and at tracts attention f rom the forecaster.
In this context, scorelines are strange, in that their salience generates utility from making precise picks of a
non-standard variable, a variable whose outcome is high ly uncertain, fragile and affected by a residual outcome that
neither team has as their preferred outcome.
1If two teams ar e equally matche d after playing e ach other twi ce, home and away, that i s the cumulativ e scoreline is a dr aw, then the team tha t has
scored more go als away from home i s the winner.
2Author cal culations usin g the entire histo ry of football m atches listed o n https://www.soc ce rbase.com , i.e. from 511,759 recorde d matches up to 8
January 2019.
Get this document and AI-powered insights with a free trial of vLex and Vincent AI
Get Started for FreeStart Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting