Scorecasting Economists

Premier League 2018/19 – Final League Table Forecast Evaluation

It could be argued that the Premier League reached new heights this season. The top two teams ended with 98 and 97 points, with just 5 losses between them all season. And in Europe, English sides completed the line-ups of both showpiece finals of the Champions League and the Europa League. RED now looks back to last August, when it made a pre-season forecast of what the final league table might look like in May, giving the most likely position of each team. A comparison of these forecasts with the final league positions of each team is below. The table also contains the % predictions the Model made for the title, Champions League qualification (top 4) and relegation (estimated using repeated simulation and random draws using the model’s probabilistic forecasts).

Overall:

As a very simple metric of forecast performance, the correlation between the August prediction of league position and the May outcome is around 0.72. This is significantly positive, implying RED has some forecasting power. It is far higher than what RED achieved for the EFL Championship. However, this is not surprising. The Championship is fairly unpredictable and has a high degree of “competitive balance”, with only some correlation in team’s performances form one season to the next. The top positions of the Premier League, however, are dominated by the same teams from one season to the next. Similarly, teams promoted from the Championship tend to do particularly poorly in their second Premier League season – colloquially known as the “Second Season Syndrome”. By most measures of competitive balance, the EPL is a fundamentally unequal league, and thus far easier to predict than the lower leagues of English football.

Title:

RED got this one right. At the start of the season the Model gave Man City a massive 92% chance of winning the title. However, during the season Liverpool pushed City close, and for large parts were ahead both in terms of actual position and RED’s predictions for the final positions.

Champions League Qualification:

Chelsea marginally over-performed this season. RED gave them a 44% chance of finishing in the top 4, with a most likely position of 6th, but in the end they came 3rd. Manchester Utd clearly under-performed this season, costing their manager his job. The Model suggested they had a 79% chance of finishing in the top 4, but eventually they came 6th and had a really poor end to the season.

Relegation:

Huddersfield confirmed expectations, finishing 20th and bottom, as predicted with an 82% chance pre-season. Brighton just about escaped the drop, which reflects their pre-season prediction of a 44% chance of returning to the Championship. Cardiff and Fulham had very small chances of relegation according to RED in August. This was the biggest failure of the Model this season. It suggests RED over-valued Cardiff’s and Fulham’s form from the previous season in the Championship and their abilities to replicate this at a higher level.

Final Pre-season forecast
Pos. Team Pos. Title % CL % Rel %
1 Manchester City 1 92.4 100 0
2 Liverpool 2 1.9 68.3 0
3 Chelsea 6 0.7 44.3 0
4 Tottenham Hotspur 3 1.1 78.3 0
5 Arsenal 6 0.2 29.9 0
6 Manchester United 2 3.7 79 0
7 Wolverhampton Wanderers 10 0 0 0.5
8 Everton 12 0 0 7.2
9 Leicester City 7 0 0.1 0.1
10 West Ham United 16 0 0 30.2
11 Watford 17 0 0 27.4
12 Crystal Palace 14 0 0 27.6
13 Newcastle United 14 0 0 22.7
14 AFC Bournemouth 15 0 0 23.9
15 Burnley 11 0 0 1.1
16 Southampton 16 0 0 32.6
17 Brighton & Hove Albion 18 0 0 44.1
18 Cardiff City 7 0 0.1 0.2
19 Fulham 9 0 0 0.4
20 Huddersfield Town 20 0 0 82

Championship – Final League Table Forecast Evaluation

As the dust settles on the English Football League season, as some teams celebrate and others lick their (financial) wounds, RED looks back to last August, when it made a pre-season forecast of what the final league table might look like in May. A comparison of these forecasts with the final league positions of each team is below. The table also contains the % predictions the Model made for the title, automatic promotion, at least playoffs and relegation (estimated using repeated simulation and random draws using the model’s probabilistic forecasts).

Overall:

As a very simple metric of forecast performance, the correlation between the August prediction of league position and the May outcome is around 0.5. This is significantly positive, implying RED has some forecasting power. But it also isn’t great in this case. The Championship though is fairly unpredictable and has a high degree of “competitive balance”, with some correlation in team’s performances form one season to the next, but not as high as in the English Premier League, for example.

Title:

RED was nowhere on this one. Norwich won the league but RED had them finishing 19th, with next to 0% chance of winning the title, and a far higher chance of relegation than making at least the playoffs. RED’s title favourites, Stoke City, really disappointed this season. RED’s forecast in August didn’t seem ridiculous, giving Stoke as much as a 46% chance of becoming champions, given the considerably greater wage budget they enjoyed compared with anybody else and their Premier League pedigree. Stoke in the end finished 16th, making a mockery of RED’s 96% prediction that they would at least make the playoffs.

Automatic Promotion and Playoffs:

Besides Norwich and Stoke, RED gave a fairly good account in August of what other teams would be in the promotion picture by the end of the season. Swansea were predicted to finish 2nd, with a 90% chance of at least making the playoffs, but in the end finished 10th. RED works on historical patterns, and in the past the teams relegated from the Premier League have generally done well the next season. However, with a large number of teams now benefiting from parachute money over several seasons, it is plausible that this relationship has weakened, or equally plausible that this season was a one-off in terms of the weak performance of the teams relegated form the EPL.

Relegation:

RED got Bolton’s demise right. Birmingham City, forecast to finish bottom, survived comfortably, which might explain why their manager Gary Monk is now being coveted widely. Ipswich propped up the final league table, but RED suggested they would finish 14th and only had a 4% pre-season chance of relegation.

The local team:

Reading finished exactly where RED said they would, 20th position. Perhaps the Model is benefiting from some sort of local knowledge or intuition, and is more in tune with its immediate environment than we thought…

Final Pre-season forecast
Pos. Team Pos. Title % AP % Poffs % Rel %
1 Norwich City 19 0 0 0.7 13.9
2 Sheffield United 9 1 2.7 31.7 0.1
3 Leeds United 11 0 0.3 5 2.4
4 West Bromwich Albion 3 11.6 28.8 83.3 0
5 Aston Villa 3 10.7 25.4 81.4 0
6 Derby County 5 5.4 15.4 68 0
7 Middlesbrough 7 1 4.4 42.1 0
8 Bristol City 5 2.4 7.6 49.2 0.1
9 Nottingham Forest 22 0 0 0.2 24.3
10 Swansea City 2 21.7 44.2 89.1 0
11 Brentford 8 0.6 2 23.1 0
12 Sheffield Wednesday 18 0 0 1.4 8.4
13 Hull City 15 0 0 6.2 2.4
14 Preston North End 12 0.1 0.2 5.2 2
15 Blackburn Rovers 14 0 0 1.8 2.9
16 Stoke City 1 45.5 68.4 96.4 0
17 Birmingham City 24 0 0 0 94
18 Wigan Athletic 16 0 0.2 3.1 3.2
19 Queens Park Rangers 22 0 0 0.4 25.2
20 Reading 20 0 0 0.2 24.1
21 Millwall 15 0 0.1 6.9 2.7
22 Rotherham United 18 0 0.2 2.4 3.3
23 Bolton Wanderers 23 0 0 0 86.6
24 Ipswich Town 14 0 0.1 2.2 4.4

Champions League Final

RED is safely back in the UK – navigating back from Baku via glorious Kiev, the scene of recent Champions League Final disappointment for Liverpool. Will it be the same tomorrow night? Will it be closer than the Europa League Final ended up being?

RED thinks so. Liverpool are favourites, without a doubt. Liverpool are 61% likely to win, Tottenham a mere 17%.

Despite this, RED thinks it will be tight – not unlike the Premier League meetings between these two clubs. But as Liverpool prevailed 2-1 in both league matches, so RED thinks a one-goal margin will suffice tomorrow night – 1-0, with probability 14%. It’s 7% likely there’ll be a repeat of the league scorelines again (2-1 Liverpool), while Tottenham are at 8% to win 1-0.

Jean-Louis Foulley agrees with RED that the reds will win 1-0 (13%), in regular time, and gives Liverpool a 64% chance of winning the Champions League overall, including extra time and penalty outcomes. There’s a 25% chance that the game will go to extra time, and a 4% chance that it’ll go all the way to penalties.

RED goes on tour

It’s the end of the season, and RED’s gone on tour – all the way to Baku!

Given RED’s at the Europa League Final this evening, it makes sense that it should make a prediction.

It’s trickier though than a league match, because we need to strip out home advantage, since both teams are thousands of miles from London.

It’s hard to escape the reality that Chelsea are slight favourites tonight. Chelsea are at 40%, Arsenal at 32% to win.

Most likely score, adjusting for home advantage? A 1-1 draw (10.8%). Conditional most likely? A narrow 1-0 Chelsea win, at 10.4%.

RED will post some more photos after the Final, although WiFi connections are hard to come by in Baku…

Championship Playoff Final, (27 May, 2019)

The richest game in football kicks off at 3pm today. Humble Derby County and heavyweights Aston Villa are facing off in the winner-takes-£170million contest at Wembley. The Model, RED, may be gearing down for the Summer but it wanted to have its say on this one. Following their end of season run of wins, Aston Villa are clear favourites, as reflected in the bookmakers’ odds too. RED says Villa have a 51% chance of a win in regular time, with Derby just 24%. The most likely outcome in the 90 mins is 1-0 to Villa (11%).

Forecast Win (%)
Score Pr (%) P(H) P(A)
Aston Villa Derby 1-0 11 51 24
  • The Forecast scoreline, with the % chance of that happening
  • The % chance of a win by the Home team, P(H), or Away team, P(A) (with one hundred minus those two numbers giving the % chance of a draw

Championship Playoffs, Second Leg (14-15 May, 2019)

The second legs of the playoff semi-finals take place tonight and tomorrow, and RED’s forecasts are below:

Forecast Win (%)
Score Pr (%) P(H) P(A)
West Brom Aston Villa 2-1 9 42 31
Leeds Derby 1-0 9 41 31
  • The Forecast scoreline, with the % chance of that happening
  • The % chance of a win by the Home team, P(H), or Away team, P(A) (with one hundred minus those two numbers giving the % chance of a draw

Man vs Machine – End of Season Premier League Results – Man can still hold head up high

In April, RED, our “super-computer” took on the BBC’s Mark Lawrenson (aka Lawro) in a football scoreline forecasting showdown. Like Deep Blue vs Kasparov in 1996, RED vs Lawro in 2019 showed that artificial forms of intelligence are a match for human intuition. Could this latest triumph for machines have been a one off? After all, even computers can get lucky.

As Lawro himself put it when confronted by the history of man being bested my machine: “It depends on the man” – confident words. But does a review of the whole season back him up?

We have been comparing Lawro’s performance with the University of Reading’s Model throughout the past season. Man can once again hold its head up high, as its standard-bearer in all things football forecasting has triumphed over the course of the whole 2018/19 Premier League season.

Playing by the rules of the BBC scoring contest, awarding 40 points for an exactly correct scoreline forecast but just 10 points if only the result is correctly picked, Lawro achieved 3,340 points for the 380 games of the 18/19 season. That is 8.79 points per game. This equates to 43 correct scores and 205 correct results. Lawro picked 1 in 9 games exactly right (11.3%) and got the result correct in 54% of games. RED achieved 3,240 points over the whole season, picking more results correctly, 210 or 55%, but couldn’t compete with Lawro’s streak of perfect scorelines, just getting 38, or 1 in 10.

Like all super-computers, RED is a work in progress, has assimilated this information, is not emotionally scarred by defeat, and will return next season upgraded to put (a particular) man in his right and proper place.

[ps. data to back this up is available on request, as well as being detailed throughout the history of this blog — Lawro’s predictions are summarised here, where in addition he made predictions of Everton 0-2 Man city and Brighton 1-0 Cardiff]

Premier League, R38 (12 May)

The final round of Premier League fixtures take place on Sunday. This is typically when fans get to see a few emerging youth prospects from their clubs’ academies, before said players spend the next few seasons on loan in the lower leagues, their prospects snuffed out by spurious summer transfer activity.

But Liverpool and Man City will certainly be playing full strength sides, as the title of Premier League Champions 2018/19 is still up for grabs. City travel away to Brighton, and RED gives them only a 73% chance of a win and certainly securing the title, with a forecast scoreline of 0-1 (15%). Liverpool have a home fixture, welcoming Wolves to the scene of their midweek miraculous and emotional triumph in the Champions League. They model suggests a 69% chance of a win, with the most likely outcome 2-0 (17%).

The table below gives the rest of the Model’s forecasts for Round 38 of the Premier League.

Forecast Win (%)
Score Pr (%) P(H) P(A)
Brighton Manchester City 0-1 15 11 73
Burnley Arsenal 1-2 9 30 42
Crystal Palace Bournemouth 2-1 9 55 21
Fulham Newcastle 0-1 14 29 43
Leicester Chelsea 0-1 9 31 41
Liverpool Wolves 2-0 17 69 13
Manchester Utd Cardiff 2-0 13 69 12
Southampton Huddersfield 1-0 18 66 14
Spurs Everton 1-0 15 48 26
Watford West Ham 1-0 10 48 26
  • The Forecast scoreline, with the % chance of that happening
  • The % chance of a win by the Home team, P(H), or Away team, P(A) (with one hundred minus those two numbers giving the % chance of a draw

Championship Playoffs, First Leg (11 May, 2019)

As Gandalf would say, “Hope is kindled” (my favourite scene from cinema). This is certainly the case for fans of four Championship clubs, who have firm hopes of their teams achieving the biggest (monetary) prize in football, promotion to the English Premier League. The first legs of the playoff semi-finals take place tomorrow, and RED’s forecasts are below:

Forecast Win (%)
Score Pr (%) P(H) P(A)
Aston Villa West Brom 2-1 9 49 25
Derby Leeds 1-0 12 47 27
  • The Forecast scoreline, with the % chance of that happening
  • The % chance of a win by the Home team, P(H), or Away team, P(A) (with one hundred minus those two numbers giving the % chance of a draw

Lower League Play Offs, First Legs (May 9-12)

The playoffs have arrived. Who can make it up still? Tight, nervy games. RED thinks all four lower league first legs will be single-goal affairs, starting tonight at Mansfield, with the hosts likely to edge their visitors Newport.

All the first leg predictions are below:

Forecast Win (%)
Score PR(%) P(H) P(A)
LG1 Sunderland Portsmouth 1-0 10% 41% 32%
LG1 Doncaster Charlton 0-1 14% 34% 38%
LG2 Tranmere F Green 1-0 13% 41% 32%
LG2 Mansfield Newport Co 1-0 15% 45% 28%