Premier Leagues 2014-2015 – Odds

Three rounds into the new Premier Leagues season and we've already seen multiple surprises. Swansea City sharing first place is one. Everton and Manchester United with two points from three matches is another. The Everton vs. Chelsea match was a true spectacle. Stoke City beat Manchester City in Manchester. Now an ever irritating interlull interrupts the early season momentum but provides time for analysis and sifting of early results to generate odds on what can be expected the rest of the season.

This summer an entertaining World Cup kept soccer fans busy. The World Cup is an interesting tournament, where pretty much anything can happen in the group stages. However, as the tournament progresses through the elimination rounds, a relentless trimming process takes place. At, we analyzed the teams in the World Cup using the ELO rating system and simmed the tournament thousands of times to produce odds for each outcome. [Links to previous articles here??] One surprising finding was the extremely high likelihood of Brazil winning the tournament (greater than 50%). Another was how quickly the weaker teams dropped out of the elimination rounds in the sims. Even a strong young team like Belgium never had greater than a 2% chance of winning the Cup.

This season the same approach will be applied to the Premier Leagues. The Premier League is a double round robin tournament of thirty eight games. The immediate conclusion is that, unlike the World Cup, we cannot expect shock surprises to occur. The season is too long and the weaker teams will never win the league. Sim the World Cup a million times and even Australia can win once or twice. Sim the Premier Leagues a million times and Burnley will never finish first and Liverpool will never be relegated.

However, even though Burnley will not win the league this season, simming based on relative team strength can still provide interesting information in three areas in particular. One, the effect of “six pointers” can be analyzed in depth. How important for their title hopes was that loss by Manchester City at home against Stoke City? Two, odds for the more cluttered milestones, like finishing in the top four, can be analyzed in depth. Three, as the season progress the effect of individual matches, for closely contested places, becomes ever greater and the odds can be compared to previous sims to see exactly what has changed.

So what do the sims show? Before the season began we had the following simulation table:

Manchester City were clear favorites, finishing first 58% of the time and averaging a 4.3 point lead over second place Chelsea. This may seem extreme but it makes sense. Chelsea did not have a strong end to the previous season, losing and drawing with Atletico, losing to Sunderland, beating Liverpool, drawing Norwich, and winning 2-1 against Cardiff. More importantly, the ELO ratings do not know that Chelsea have strengthened over the summer with Cesc Fabregas and Diego Costa, amongst others. ELO is based purely on results.

Liverpool finishes a clear third in the pre-season projections. Remember, ELO does not know about Suarez leaving or all the new players coming in. Results will tell the tale. I suspect Liverpool is rated a bit too highly and will decline over the season.

Manchester United, despite finishing in seventh last season, is projected to finish fifth. The projected top seven is in a class by themselves, with a 14 point difference between Tottenham and Southampton. With all the personnel changes I suspect Southampton will be worse than projected and Manchester United slightly better.

Surprisingly, Hull is predicted for relegation. This is likely overdone for three reasons. One, Hull finished quite poorly last season, going 3-9-1 in their last ten matches after appearing “safe” earlier. Two, they did not do well in the Europa league against competition that, according to ELO ratings, was significantly worse than they were. This is strongly penalized in ELO. Three, the newly promoted sides are almost certainly stronger in rating than they are in reality, since they were playing one division lower last season.. This will adjust itself after five to ten matches in the Premier Leagues. Look for Hull to improve by late October.

Pre-season predictions are much less interesting after the season has already started. Let's get to the good stuff. Three matches in and what has changed? Chelsea has looked strong. Manchester City lost, at home, to Stoke. Manchester United has continued to under perform their ratings. Everton only has two points. Swansea has played excellent soccer and sits joint top. Aston Villa, with a 25% chance of relegation in the pre-season sims are in third place with 7 points. Here is the new table.

The first obvious change is that Chelsea is now predicted to finish first. The loss against Stoke at home by Manchester City dropped their odds of finishing first from 70% to 39%! Early season matches matter just as much as late season ones and this was a massive upset. The loss being at home makes it worse, as the simulations expected Manchester City to win 75% of the time. Amazing that one loss can make such a difference. The chances of a tight race have increased.

Liverpool also benefited from the Manchester City loss and their own dominating win away at Tottenham. Their odds of finishing first increased from 6% to 14% and finishing top four from 86% to 96%.

Manchester United? There are two bright spots in their horrid start to the season. One, they have been busy in the transfer market, getting Di Maria and Falcao. Two, their loss to MK Dons does not count against their ELO rating because, the source of ELO ratings for club level soccer, does not include cup matches because of the large lineup changes. Otherwise it is grim. Manchester United was predicted to have seven points after three matches, against weaker opposition, and now they have two. The sims now predict they have almost no chance of finishing first and only a 16% chance of finishing top four. After three matches they are now dependent on other teams playing poorly above them.

Everton have also under-performed, having 3.4 less points than expected.

The biggest surprise of the season so far has been Swansea. They are now expected to finish with 6.6 more points than they were originally predicted to have. They also have a 0.44% chance of finishing top four. I don't believe that will happen either but what a great start for Swansea. Top seven is not impossible for them.

Aston Villa have started very well. Their seven points decrease their odds of relegation from 25% to 8%. Leicester City still is predicted to have a lower chance of relegation than Aston Villa but, as mentioned above, the promoted teams are almost certainly rated too highly. Sure enough, three matches in and all three promoted teams have seen their predicted point total decline. This adjustment process is likely to continue for the next few weeks.

Of the relegation targets, the worst performing team so far this season is Crystal Palace. Last year's plucky over-acheivers under Tony Pulis were probably due for a reversion to the mean this year regardless. Their start has been poor. Already they are expected to have 4.5 points less than the original sims indicated and their relegation odds have increased from 26% to 40%.

Hull's expected improvement has not occurred. Although they have four points, the poor performances in the Europa league really hurt their overall rating. I still expect a rebound.

QPR has helped their cause with three points and a decrease in relegation odds from 75% to 69%. Still, they are averaging only 28.7 points a sim. Without significant improvement over the next five matches they would be easy favorites to go down.