As a football (soccer) fan, one aspect of American sports that I’m a bit jealous of is the draft. Not because I want bad teams to be rewarded – that part is ridiculous – but I think the annual speculation would be a lot of fun. Who would have been the top picks in years gone by? Who would be the huge busts and the undrafted superstars? And how much would a high draft pick be worth in trades?

For the last decade, goal.com has been publishing its NXGN list containing the top 50 players aged 18 or younger on the 1st of January. This is just their subjective ranking, of course, but it does give us a snapshot of how each player was seen at the time rather than with the benefit of hindsight.

(Note that some players can be aged 19 by the time the list is published around March).

It’s a mixed list containing superstars (Yamal, Bellingham, Donnarumma), disappointments (Fati, Sancho) and some that fall in-between (Rodrygo, Tielemans).

Ignoring the specifics of how an actual global draft would work, let’s pretend that the NXGN list represents the first 50 picks in a draft each year. 18 seems like the right age for a football draft – players are still largely just young prospects with potential rather than established first-teamers at the top level.

To get a sense of the value of each pick I used the last published Transfermarkt values prior to each year’s publication. These were inflation-adjusted to 2026 equivalents by taking the median of the top 500 Transfermarkt values each year and fitting a cublic smoothing spline to the logarithm of the medians.

Below we can see the median and mean value when we group picks 1-50 into various bands. Immediately we can see that a top 5 pick is worth dramatically more than lower picks.

But those are the valuations at the time the list was published – it would be reasonable to think NXGN’s order will just reflect the same consensus used to generate Transfermarkt values. What we are really interested in for these young players is their future value. The following plot compares two curves: one models the value of each pick at time of publication, and the other models the value of the player five years later. To emphasise, these values are all normalised to 2026 levels, so an increase doesn’t simply reflect inflation.

It appears that a top-5 pick is very valuable, with a forecast value of more than €125m after five years. After the top 5 the value drops off quite quickly, and beyond the 15th pick the curve is relatively flat – having the 40th pick isn’t worth much more than the 20th pick.

It’s also noticeable that players at all levels are expected to be more valuable after five years. An obvious conclusion would be that young players are undervalued, so just buy all the young players! But this misses an important point: football recruitment is not a free market. Smaller clubs do not have the leverage to simply hold onto a player forever, and among buying clubs there is often a “preferred bidder” that the player wishes to join, or a small group of rich & successful clubs that would be considered. So the big clubs are in a strong position and can sign the best young talent for relatively good value.

Which picks were the biggest busts?

• Reinier Jesus (#4 pick in 2020). Just after his 18th birthday, signed with Real Madrid for €30 million. Bounced around Europe on various loans before moving back to Brazil on a free transfer in 2025. Currently valued at only €2m.

• Gio Reyna (#3 pick in 2021). Another great American hope who broke into the Dortmund side at the age of 17. Struggled to follow that up and after a disappointing season with Borussia Monchengladbach he has just joined Strasbourg for the grand fee of €3m.

• Ansu Fati (#2 pick in 2020, #1 pick in 2021). Seemingly the next Barcelona phenom when he debuted for the club at 16 years of age. Injuries have blighted his career and he recently joined Monaco for €11m after a mediocre loan season.

• Alban Lafont (#4 pick in 2017). Became a regular starter in Ligue 1 at the age of 16, which is absurdly young for a goalkeeper. Has had a respectable career with more than 300 starts in Ligue 1 and Serie A, as well as 4 caps for the Ivory Coast, but never kicked on to the anticipated levels. Now playing in Turkey with Amedspor.

• Other notables: Justin Kluivert (#1 in 2018), Callum Hudson-Odoi (#3 in 2019), Breel Embolo (#2 in 2016) and Mason Greenwood (#3 in 2020). Not dreadful players by any means, but none of them reached the anticipated heights for a variety of reasons.

Which low picks were steals?

• Erling Haaland (#33 pick in 2019). It feels like someone with Haaland’s physical tools was always an inevitable superstar, but NXGN ranked him well down the list in 33rd, just behind Rhian Brewster and Emile Smith-Rowe.

• Bukayo Saka (#15 pick in 2020). Saka was starting regularly for Unai Emery’s Arsenal at 18, which nowadays would mark you as an elite prospect, but it was a poor Arsenal side which perhaps dampened the hype a bit.

• Julian Alvarez (#37 pick in 2019). Alvarez was a relatively late bloomer, not establishing himself in the River Plate XI until the age of 20. Even at Manchester City he had to play second fiddle to Haaland, and it is only now that he is being viewed on the elite level as he tries to force a move away from Atletico Madrid.

Undrafted players

There are also a large number of valuable players who didn’t make the top 50, reinforcing the idea that beyond the top handful of prospects player-development is a bit of a lottery. Here are the most valuable undrafted players (using their value five years later) from each of the NXGN editions from 2016 to 2021. I restricted this to players aged 18, because it’s a bit harsh to expect them to include someone at age 15 who became a star player by 20.

It’s surprising to see that players like Frenkie De Jong, Declan Rice or Joao Felix did not make the top 50 because they all feel like they emerged at a young age. But it just shows, when we strip away the benefit of hindsight, it is not inevitable which 18-year-olds will be the next generation’s best players. Felix is especially surprising because by the age of 19 he had already broken into the Benfica team and been transferred to Atletico Madrid for €126 million! Things can develop very quickly.

What about actual transfers of young prospects?

As mentioned earlier, the NXGN ranking is just a subjective list. Let’s take a look at some actual players who were bought for large fees before the age of 19, and whether their Transfermarkt value five years later was higher or lower than the transfer fee.

It’s a mixed bag: of the 20 players shown, half decreased in value and half increased. In fairness, Oxlade-Chamberlain and Pjanic barely decreased, while some of the increases were very large, so it does seem like the 20 purchases were successful in the aggregate.

The most expensive 18-year-old was Luke Shaw whose move from Southampton to Manchester United is equivalent to more than €100m in 2026. Shaw has had a respectable career: over 200 appearances for United and 34 caps for England. But he has struggled with injuries and it’s not the return you hope for when investing such a huge amount in a left-back.

The NXGN prospects list only dates back to 2016 but if we use Transfermarkt values to determine our “pick” order, we can go back a little further. We can also assess who the top pick would have been each year, not only in a global draft, but also in a draft among England-based players only.

Surprisingly, Luke Shaw is the only England-based player since 2009 who has been the most valuable 18-year-old in the world. The overall list of English #1 picks is underwhelming; Shaw is actually one of the more successful outcomes compared to the other youngsters, some of whom didn’t even establish themselves at the top level.

One prominent aspect of this chart is just what an outlier Lamine Yamal is. As the Messi & Ronaldo era comes to an end, it seems nicely set up for Yamal to take over as the world’s clear superstar. It’s also notable that Chelsea’s Estevao is the most valuable young prospect in English football for some time; at time of writing it’s not easy to see where he will fit into Xabi Alonso’s formation, but talent has a way of finding its way to the top and it will be a surprise if Chelsea decide to let him go.

Takeaways

So it seems like if you can get one of the top 5 prospects in world football, great. You are probably onto a winner. Beyond that, it becomes a bit more of a lottery: just get as many tickets as you can, and see which ones win.

NXGN’s 2026 top five are therefore interesting to watch. Email your club, and tell them to snap these players up! Except, bad news…the big clubs already have them:

“Here we go!” You’ve got to admit it’s a good catchphrase. Transfers are big business in modern football, and transfer news and rumours have become entire careers for some personalities. But how much do transfers really affect future performance? And what is the best way to measure that?

I looked at data from the fantastic website Transfermarkt to compare all transfers in the Premier League since 1992/93. The first challenge was to adjust all transfers for inflation. There are different ways to approach this, and they are all fairly arbitrary, but I normalized each transfer as a proportion of the total Premier League signings in that season and then added some smoothing in case there were outlier seasons that skewed the results. The top 10 biggest inflation-adjusted signings by Premier League clubs were as follows:

It’s surprising for Rio Ferdinand (£228m adjusted for inflation) to come out on top, not least because he is a defender. I remember thinking at the time it was a crazy fee (£29m from Leeds in 2002, rising to £33m with add-ons) but he spent the following decade as a mainstay of the Manchester United and England defence, so I think we can safely say Alex Ferguson is a better judge of a player than I am.

Despite the huge volumes flying around the Premier League in recent years, it is notable that only one transfer in the last 15 seasons (Paul Pogba to Man Utd for inflation-adjusted £160m) makes the top 10. Why is this? I think it’s fair to say there haven’t been any seismic transfers for quite a while. The last couple of summers are a good example: big-money moves have seen Alexander Isak join Liverpool, Elliot Anderson join Manchester City and Morgan Rogers join Chelsea. But these are all typical moves: the best player from a moderate-sized club to a bigger one.

To imagine a modern equivalent of Fernando Torres joining Chelsea (£200m inflation-adjusted), you would have to consider a star forward from a Big Six club forcing a move to another Big Six club. Perhaps Bukayo Saka to Man City — how much would that transfer fee be? And Alan Shearer (£185m inflation-adjusted) was unquestionably the league’s best goalscorer when he moved to Newcastle at age 26. What price Erling Haaland if he moved to another Premier League club this summer?

I’m sure many of you are curious how other big signings stack up when adjusted for inflation. Below is a table showing the biggest adjusted fee paid by each of the 51 clubs who have competed in the Premier League. This will be a head-in-hands moment for fans of certain teams.

The arrival of Roman Abramovich into English football was truly a game-changer. In Abramovich’s first two seasons, Chelsea’s net spend was equivalent to about £1.5 billion today. The table below shows the top 10 net-spend seasons by Premier League clubs, with Chelsea & Abu Dhabi-owned Manchester City occupying all of the top 8 spots.

Given the obscene levels of spending, it’s no surprise that Chelsea and Man City have hoovered up a lot of trophies. I’m not going to pull up any trees by concluding that spending leads to success, but let’s try drilling down a bit to see how strongly various spending measures correlate with future team performance.

Is net spend the best measure?

An obvious starting point is comparing each team’s net spend in the summer transfer window with their performance that season, as measured by points-per-game (PPG). Prior to the window’s introduction in 2002, I have used the 1st of September as the cut-off.

As expected, higher net spend is associated with higher PPG, with a Pearson correlation coefficient of 0.26. But is net spend a useful metric? What if we take a separate look at the total fees paid and received by each club? Also, rather than just looking at summer spending, I considered the summer window plus the previous two seasons’ transfer activity to account for the overall investment in the team rather than just the most recent window. The correlation coefficients for the six combinations of these two variables can be seen below.

Unsurprisingly, correlation was improved by including the previous two seasons. But we can also see that net spend isn’t necessarily the best metric; the strongest correlation was found using total transfer fees paid, disregarding fees received altogether.

There is also the question of whether the cumulative sum of fees is the best metric; it is possible to sign a large quantity of players who don’t improve the team. What if we instead look at the biggest fee paid?

That’s an interesting result. The strongest correlation (r = 0.64) to teams’ performance is found by using the biggest transfer fee paid in the summer and the prior two seasons – a larger correlation than with total fees paid. There is some intuition to this result: at any point in time, the biggest transfer fees will generally be paid by the biggest clubs. Whereas, total fees paid and received can be confounded by other aspects of team performance. For example, a team buying a lot of players may be doing so because they have been underperforming and need upgrades, while a team making one or two expensive sales may be doing so off the back of good team performance. So it’s not clear which direction the causal arrow would point.

Controlling for team effects

Of course, looking at overall relationships between transfer spending and performance just leads to a conclusion that a seven-year-old could figure out: rich clubs like Manchester City spend more on players than small clubs like Luton, and therefore they get better results. In addition, clubs go through good and bad cycles, so we can’t just assume each season’s performance is entirely down to recent transfer activity.

To control for these factors, I trained multiple linear regression models with the original transfer features and additional features containing the team as a categorical variable, plus their prior season PPG and goal difference per game (GDPG).

I accounted for promoted teams by training a separate model for the effect of going from the Championship to the Premier League which also included a time variable; as hypothesized, the time variable had a negative slope because the gulf between the Championship and Premier League has increased through the years.

The coefficients from the multiple linear regressons are shown below. The farther the dot is to the right, the more PPG increases as that variable increases.

The strongest predictor of PPG was total fees paid over that summer and the two previous seasons, followed by net spend over the same period. Both the total fees received in the summer window, as well as the size of the biggest fee received in the summer window, had a small negative correlation with the next season’s PPG.

Finally, we can deconstruct the summer window plus two previous seasons into its constituent parts to see how each correlates with PPG (see below). When it comes to fees paid (the plot on the left), in the current summer the biggest fee carries more weight than the cumulative total, whereas it’s the opposite the season before: the total spent is more important. In fact, the total amount of fees spent across the previous season seems to be the most powerful single measure. Going back two seasons, the effects are close to zero.

As for transfer fees received (the plot on the right), the effects for both the cumulative total and the biggest fee are similar. In the current season’s summer window, larger player sales have a small negative correlation with PPG. One season prior, the effect is close to zero, and two seasons prior it is a small positive effect. It seems like selling players for decent fees isn’t an especially bad thing, but perhaps it is some short-term pain for long-term gain, which makes sense.

Conclusions & limitations

To summarize the main findings:

• Total fees paid are a better predictor of performance than net spend, with total fees paid over the summer plus the previous two seasons the strongest overall predictor.

• Fees received in the current summer window have a small negative correlation with performance.

• Total fees paid in the previous season are a better predictor of performance than fees paid in the summer window.

This study has some obvious limitations. Wages are not included, because frankly I don’t know an easy and reliable source for tracking historical player wages. Transfer fees on Transfermarkt are not gospel – some fees are undisclosed, estimated, or may include add-ons. In the modern era the loan-to-buy deal is becoming more commonplace, which means the value on the books may relate to a player who was already at the club the season before. My inflation measure is arbitrary; different methods may produce different results. Fees paid by promoted teams in a lower division may be falsely low because of the Premier League premium paid by top flight teams. And lastly, because spending and performance are interrelated, causality is hard to assess.

That said, I think there are some interesting high-level results and we can see how they pertain to teams in this year’s Premier League. Chelsea are comfortably the biggest spenders over the summer plus the last two seasons with £860m invested, while Liverpool fans will be hoping that the one-year lag benefits them this season after their massive £412m outlay on players last year.

As for negative indicators, both Newcastle and Aston Villa have received north of £200m in player sales this summer, each with a single player going for £100m or more. The good news is, if they can withstand the short-term hit this season, their long-term prognosis may not be too bad.

In last week’s article I looked at which statistics were the best predictors of future performance in the Premier League. One of the comments was from porterhouse26 who asked about modelling promoted teams, so I thought I would take a look at that along with doing the same for relegated teams and the Championship.

Again we will start with a correlation matrix looking how each stat correlates with each other stat from one Championship season to the next; the black-outlined cells show how stats correlate with themselves.

We can immediately see that there is less correlation from one year to the next in the Championship than in the Premier League. The highest correlation is once again found in possession, but it is only 0.67 for Championship teams compared to 0.88 in the Premier League. In fact the lowest correlation anywhere in the Premier League grid was 0.68, showing that the Premier League is a lot more stable and predictable than the Championship is.

Similar to the Premier League, the univariate regressions find xG and touches in the box to be the best predictors of the following year’s points-per-game (PPG) in the Championship. However, in the three multivariable models shown below, xG and possession are the two strongest predictors.

I wouldn’t read too much into this because there is a lot of overlap between the variables. As a general rule it seems like if you include some measure of shot quality (goals or xG) and some measure of dominance (touches in the box or possession) then you are most of the way there.

Using an average of these three multivariable models, we can forecast the PPG for all Championship teams in previous seasons. Then, for those who got promoted, we can look at the difference between their PPG in the Premier League and the PPG forecast by the model, and calculate the average effect of promotion. We can also do the same for relegated Premier League teams using last week’s models. The net result is that relegated teams are expected to improve by 0.71 PPG and promoted teams are expected to get worse by 0.77 PPG.

Using these models and the baseline adjustments, we can forecast the number of points for the three promoted teams next season. Hull City fans may wish to look away now.

Why are Hull (15 points) rated so much worse than Coventry City (38) and Ipswich Town (38)? Let’s compare the stats of all the promoted Championship teams going back to 2016/17.

Hull’s stats really are astonishingly poor for a promoted side. They are the only team with a negative xG difference (-0.33 per game) and less than 50% of touches in the box in their games. This model forecasts them only 14.7 points; the second-lowest projection was Cardiff City with 25.8 points in 2018/19.

One question I raised in my reply to porterhouse26 was whether certain types of team are better-suited to doing well after promotion. I looked at four stats covering two characteristics:

1. How attack-minded is the team? (Total Goals, Total xG)

2. Do they like to dominate the ball & territory? (Possession %, Touches in the Box %)

Each of these variables was added on its own to the existing model to see how it changed the PPG forecast.

All four variables were associated with a decrease in PPG. The largest decrease was in teams with high possession %, and this was the only variable that improved the existing forecast when added to the model. This result is intuitive, because a team that is built to dominate possession against inferior teams may have to learn a completely new playing style when facing superior opposition.

I’m always wary of overfitting, so I used a nested cross-validation to find a shrinkage-weighted coefficient to include this additional possession adjustment in an enhanced model version. Let’s see how this adjustment affects the three teams’ projections:

The possession-enhanced model drastically boosts Hull’s projection from 15 points up to 24, which is about where the betting markets have them. Ironically, the fact Hull were so bad last year may work in their favour to some extent because they are used to defending!

We can also see how the possession adjustment changes the forecasts for all the promoted teams in previous seasons:

It’s clear that some of the most possession-heavy Championship teams had disappointing seasons after getting promoted. The two highest possession teams were Russell Martin’s Southampton (61.9%) who got just 12 points, and Vincent Kompany’s Burnley (61.1%) who got 24 points. In fact, of the top 10 promoted teams with the highest possession % in the Championship, only Leeds in 2020/21 and Fulham in 2022/23 outperformed their baseline forecast.

At the opposite end, recent evidence can point to Sunderland who got promoted through the playoffs with just 49.4% average possession across the Championship season then stunned everyone by qualifying for the Europa League with 54 points.

What about relegated teams? Is there a type of relegated team that tends to perform well in the Championship?

The evidence would suggest not. The coefficients are much smaller than those for promoted teams, and none of these variables improves the model. This suggests that multiple playing styles can work well in the Championship. Or, more simply, it could be because relegated teams quite often change manager after a disappointing season so their style may be different anyway, whereas promoted teams generally stick with the same manager unless he is poached by a bigger club.

Given that the playing style enhancements don’t add any value, we can use our baseline model to project next year’s Championship, including the three relegated teams. The teams promoted from League One are missing because I don’t yet have League One in my database, but I’m working on that.

I don’t follow the Championship as closely as I follow the Premier League, so I’m not going to get into the weeds of why some of these teams have a projection higher or lower than the betting markets. Middlesbrough immediately leap out; one might expect that any team that is overprojected by the model has perhaps been weakened in the summer transfer market, but in Boro’s case it seems to be the opposite. According to Transfermarkt they are the Championship’s third-biggest biggest net spenders, headlined by the arrival of centre-forward Will Lankshear from Spurs. Boro seem like a solid buy at 76.5 points and decent value at 10/1 to win the division.

Two other teams where the model is bullish are West Brom and Watford. I’m not intending this newsletter to start being a tipping service, and I have no idea about these teams to be honest; I think it’s just interesting to take note and look for reasons in the disparity. I’d be interested if anyone who follows the Championship more closely than me can explain why the markets might differ from my model on these two teams.

Conversely, Wrexham are hugely underrated by the model relative to the spreads. This could be due to a public overrating of the Hollywood-owned club, but they have some wealthy backers and one of my maxims is you should try not to swim against the tide. It seems much more likely that Wrexham’s next division change is upwards rather than downwards.

West Ham feel like a squad that are miles too good for this division but these can often be trappy mug bets and this is where a numeric model-driven projection is really good at giving you discipline. Are West Ham right to be sub-2/1 favourites for the Championship? Debatable. Could you justify backing them because you think they should be even shorter than that? It feels like a big leap, and while they have done well to clear up a large proportion of their financial deficit via the big-money sales of Mateus Fernandes and Crysencio Summerville, let’s not ignore the fact that those two players may have been a large reason why West Ham’s attack was pretty good in the latter half of last season. Not to mention, the window hasn’t closed yet and it would be no surprise if the impressive Taty Castellanos has departed by the end of August.

The saying “all models are wrong, but some are useful” is often heard in statistical circles and it really is a phrase worth taking to heart. None of the projections in this article should be taken too literally because there are a ton of factors not accounted for (e.g. summer transfers). But often the most important part of modelling is not so much the projection but understanding how the model got to that projection, and then using your domain knowledge to assess whether or not that’s something the wider market may have missed. That’s what makes a model useful.

The new Premier League season is still a couple of weeks away but on Wednesday night we have an hors d’oeuvre as Paris Saint-Germain take on Aston Villa in the UEFA Super Cup at the Red Bull Arena in Salzburg.

Does anyone care about the Super Cup? It’s obviously nice to be in it by virtue of having won a European competition, and it’s similar to the Community Shield: if you win then it’s important, if you lose it’s just a glorified friendly; and if you are absolutely desperate to win an argument about trophy counts then you will chuck it in as a proper trophy.

Originally the Super Cup was between the winners of the European Cup and the European Cup Winners’ Cup, prior to the demise of the latter when it was replaced by the winners of the UEFA Cup. Until 1998 the Super Cup was contested over two legs, although due to scheduling difficulties it was occasionally played over a single leg (1984, 1986, 1991) and sometimes not at all (1974, 1981, 1985).

The below chart shows for each iteration of the Super Cup the difference in Elo rating between the European Cup/Champions League winners and their opponents (the winners of the Cup Winners’ Cup/UEFA Cup/Europa League).

It is clear from this chart that the Super Cup has become more one-sided over time. The last time the Champions League winner was the underdog according to Elo ratings was in 2007 when European champions AC Milan played Sevilla. It may seem strange that the Milan side of this era were rated lower than Sevilla, but Milan had only finished 4th in Serie A which was without Juventus who had been relegated due to the Calciopoli scandal. Sevilla, meanwhile, had finished third behind Real Madrid and Barcelona, and only five points off the La Liga title.

Although PSG are considered strong favourites over Aston Villa, according to Elo this is actually the closest matchup in a Super Cup since 2012 when fortuitous Champions League winners Chelsea were thumped 4-1 by Atletico Madrid.

Ironically, the year that the European champions were the biggest-ever underdogs was 1982 when it was Aston Villa themselves who were rated 121.9 points worse than Barcelona — nonetheless, Villa prevailed in extra time after a 1-1 aggregate score over two legs.

Of course, the difference between the two Super Cup teams only tells part of the story. The following scatterplot compares the ratings of the teams in each iteration so we can see the strongest and weakest sides down the years.

As you can see, the Aston Villa side of 1982 are the lowest-rated European champions to play in the Super Cup. Villa had only finished 10th in the previous season’s English Division One, still a record low position for any European champions. The lowest combined rating was in 1980 when Nottingham Forest, 5th in Division One, faced Copa Del Rey winners Valencia who had finished 6th in La Liga.

The nine strongest European champions according to Elo ratings are all since 2011, demonstrating the increased concentration of resources among a small number of elite clubs. The 2011 clash between Barcelona and Porto is rated as the strongest-ever Super Cup in terms of the combined Elo rating, while the worst-ever Elo rating for a Super Cup team was 1995’s Real Zaragoza who famously won the Cup Winners’ Cup against Arsenal with a last-minute goal from near the half-way line scored by former Spurs midfielder Nayim.

Only two English teams have won the Super Cup more than once (Liverpool four times; Chelsea twice) and Villa will be looking to join that company this week. Unai Emery’s prowess in European competition is well-documented, with an astonishing five Europa League triumphs. Unfortunately he is so far winless in three attempts at the Super Cup but after a gruelling World Cup and still another 10 days until the season starts, it would be no great surprise if this game is a bigger priority for Villa than for PSG and Emery finally gets off the mark in this competition.

I looked at Premier League seasons going back to 2014/15 and used seven stats from each season (points-per-game, goal difference, xG difference, shots %, corners %, touches in the box %, and possession %) to predict the following season’s points for each team.

Firstly, it is always interesting to create a correlation matrix to see how each stat correlates with each other stat (including with itself). In the correlation matrix below, the cells along the diagonal are marked with black borders — these are the self-correlation scores that show how well each stat predicts itself in the following season.

Interestingly, the stats that predict themselves most reliably are possession and touches in the box, which may both reflect style of play rather than just team quality. And it is notable that the stat that is least predictive of itself is points per game!

A starting point to measure the effect of each statistic is by running a separate univariate regression using each of the seven measures on its own to predict teams’ points-per-game in the following season.

Interestingly, what we see here is that any individual metric is going to give you a roughly similar level of performance. Most people are aware that xG is a more useful measure than a team’s share of corners; and yet, if you only knew the share of corners for each game, you still wouldn’t be far off! There is a lot of crossover between all these stats because ultimately, if you are dominating a match then you will dominate all these stats.

What is more interesting is when we apply all these stats simultaneously to predict the following season’s points. There are different ways to do this within a regression framework; I tried three methods:

• Multiple linear regression

• Ridge regression

• Stepwise feature selection

A multiple regression is where all variables are chucked in and a best fit is achieved. This is the simplest approach but can involve problems of multicollinearity: if two variables carry the same signal then you can get funny results where one of them may have a negative coefficient even though it has a positive association with the response variable. We can see this effect below — clearly each team’s points would not be a negative predictor of their points the next year, but that is how the model comes out because points are so heavily related to all the other predictors.

Interestingly, goals and touches in the box are the two dominant predictors. One might have expected xG to be top, but xG has crossover with both goals and touches in the box (a good reflection of territorial dominance) and once you have included those two measures, xG contains less additional value.

Ridge regression involves a penalty that shrinks coefficients towards zero, which reduces overfitting and is better at handling correlated predictors. The results of the ridge regression can be seen below.

The results of the ridge regression are more smooth, with each of the seven predictors contributing a reasonable chunk of the final estimate. Touches in the box come out on top, followed by xG and then goals. There is no doubt that these are the “big three” of simple football stats when you want a general idea of team quality.

The final regression used stepwise feature selection. This is where you add and remove variables to and from the model and see which combination gives the best results. In this context, “best” can be judged in various ways, but I used something called the Akaike Information Criterion (AIC). Interestingly, the optimal model was very simple, using just touches in the box and goals.

Now that we have these models, the natural question is how would they forecast the approaching 2026/27 season? And how do those forecasts compare to the betting markets? I looked at the current season point spreads from Sporting Index. For the uninitiated, spread betting involves a variable risk where you buy or sell at a price and the amount won or lost will vary based on the final result. So for example if you buy Arsenal’s points at 80 for £100 a point and they finish the season with 70 points, you lose £1,000.

Below are the mean projections for each team using the three models. Despite the different methods, the forecast for each team does not differ greatly. Again this shows the amount of crossover between all the stats.

A couple of things stand out. Firstly, the points projections seem far more congested than a typical league table. It would be a mistake to interpret these projections as the league champion getting 75 points and the worst of the teams getting 45 points. These are simply the mean projections for each team; in reality, once variance kicks in, some teams will drastically overperform and others will underperform, and the overall spread will be greater. But it is also an era where the league is far more congested than previously; Arsenal were far less dominant than previous league champions, and the fact Spurs almost got relegated is a demonstration of the Premier League’s current depth.

The numbers in parentheses denote the difference between the models’ projections and Sporting Index’s point spreads. One thing that leaps out is that the big six teams (Manchester City, Arsenal, Liverpool, Manchester United, Chelsea, and Tottenham) are all forecast to be worse than the betting spread.

The primary reason for this is that my models omit some important information. The wealthiest teams are more likely to improve (or less likely to get worse) from one season to the next, because a) they tend to spend money recruiting better players each summer and b) an underperformance is more likely to be due to reasons such as injury crises or bad management rather than lack of talent.

Nonetheless, I think we can look at the disparities and assess some examples where the difference is fair and where it may be challenged. The biggest negative differences between my models and the spreads are Chelsea (-10), Tottenham (-10), Manchester United (-9) and Liverpool (-8).

In the case of Chelsea and Spurs, I believe there are mitigating circumstances. Chelsea were looking reasonably good before Christmas, then suffered from Enzo Maresca’s resignation and the dubious appointments of Liam Rosenior and then the interim manager Calum McFarlane. Their season really tailed off when they were arguably the third-best team in the league. And Spurs’ problems were well-documented: an astronomical injury list along with one poor manager (Thomas Frank) and one diabolically bad one (Igor Tudor). But the owners have opened the purse strings and a big improvement under Roberto De Zerbi seems likely. It is also a huge advantage for both of these London clubs that they have no European football this season.

Conversely, Manchester United had no Europe last season but return to the Champions League slog this year. They had a fair amount of things go their way under Michael Carrick and benefitted from career years for Casemiro (now departed) and player of the year Bruno Fernandes, who is turning 32 next month.

Liverpool sacked Arne Slot due to last season’s underperformance but many of the issues may run deeper than the head coach. Between Mohamed Salah, Ibrahima Konate and Andy Robertson, plus Curtis Jones who by all accounts is likely to depart this month, they have lost more than 8,000 league minutes from last season. Cody Gakpo has also been linked with a move to Spurs and their defensive rock Virgil van Dijk is now 35 years old, while Hugo Ekitike’s impressive debut season unfortunately came to an end with a long-term Achilles injury that will keep him out for much of this season.

Given these factors, it seems like the arguments for selling Liverpool and Man Utd points at 71 and 68 respectively are more solid than those against Chelsea & Spurs.

What about the teams predicted to overperform the spreads? Many of these teams have been negatively affected by player sales (Newcastle, Crystal Palace, Nottingham Forest) and/or manager departures (Newcastle, Bournemouth, Fulham, Crystal Palace). A couple of teams that look like solid buys to me are:

• Brighton at 52. I repeated ad nauseam at the back end of last season that Brighton had been really good after re-signing Pascal Gross in the January window. They lost Jan Paul van Hecke to Spurs, but have brought in Pascal Struijk and Luka Vuskovic, and it feels inevitable that some of Brighton’s young talent will take a leap forward this season. They are just so well run, and I think one of these years they will suddenly have an unexpected monster season and qualify for the Champions League. Maybe this is the year.

• Leeds at 48.5. Leeds were good last year, even though they ended up embroiled in the relegation battle until fairly late on. Daniel Farke made a key system change at half time away to Manchester City in November, and from that point on they played like a top-half side. Tellingly, goalkeeper James Trafford chose Leeds over Newcastle this summer, and that sort of thing can be a decent indicator of which direction clubs are heading. Leeds have wealthy backers, they are a Premier League-sized club and they are here to stay. I can see them finishing in the top half and maybe qualifying for a European spot.

Including historical team finances and summer transfers would be a good way to improve on these models and I’ll look into adding that if I can get the data. In the meantime, I would love to hear any views or feedback. Which teams do you think will over- or under-perform the spreads this year, and why?

It is hard to escape the current ubiquity of the prediction market platform Kalshi in the United States. Given my extensive background with the world’s leading betting exchange, Betfair, I was naturally curious how Kalshi’s exchange model compares.

Betfair’s arrival on the U.K. betting scene at the turn of the century was truly a game-changer for punters. Most obviously, the novel peer-to-peer format allowed better prices, trading in and out of positions, and more extensive in-running markets. But also, and perhaps more importantly, there actually existed the prospect that you could be a winning punter and not have your account promptly closed, as is the case with high street bookies like William Hill or Ladbrokes.

The American betting landscape is a lot more complicated than Britain’s, with piecemeal laws existing across various states due to moral objections, political lobbying and vested interests. In fact, Kalshi just got sued by the state of New York for running an illegal gambling operation. I’m not getting into all that stuff because I’m ultimately a numbers man; what I am interested in is how does Kalshi compare to Betfair in terms of pricing, liquidity and bet-matching?

The first thing to note is that while Kalshi and Betfair are both betting exchanges, Kalshi frames its bets as contracts. I assume this is done to maintain the pretence that a prediction market is more like a financial market than sports betting, even when the contract is on whether or not a football team will win a match. Each contract pays out $1 if the bet wins, and the price of the contract is what is being agreed between the buyer and seller. Simple example: a 50¢ contract means the buyer pays 50 cents for the contract and doubles his money if it wins. So it is easiest to frame this in percentage terms: buying a 50¢ contract means you think the probability is greater than 50%, and selling the contract means you think it is lower than 50%.

Originally, Kalshi offered all markets in price increments of 1¢ (51¢, 52¢, 53¢, etc). At some point, they decided to offer certain larger markets, like the World Cup, in smaller increments (51.1¢, 51.2¢, 51.3¢, etc).

A quick glance at any major market on Kalshi shows that liquidity isn’t a problem and gross prices (i.e. before commission) are good. For example, I randomly opened a live baseball game between Philadelphia and Miami and saw that you can buy or sell Philly at 58¢ or 57¢ for large volume. This is equivalent to a spread of 1.724 - 1.754 in decimal prices, or a 101% book.

The question is of course, how much do you then have to pay in commission on winning bets? On Betfair, you would have to pay 2% of your net profit per market. What about Kalshi? The answer is, it depends!

Kalshi uses different rates of commission — or fees as they call them — depending on whether you are a market-maker or a price-taker (I will henceforth refer to these as ‘maker’ and ‘taker’). Whenever a bet is matched between two parties on an exchange, naturally one party was in the market first with their order waiting to be matched (the maker), and the other party subsequently matched it (the taker).

The fees are calculated using the formulae

where C = number of contracts and P = price in decimal form i.e. 50¢ = 0.5.

The formulae for calculating the maker’s fee and the taker’s fee look complicated but they actually aren’t. On Kalshi you pay the fee when the bet is matched, so you are paying commission whether you win or lose. But the P(1 − P) part of the formula ensures that your fee reflects both the amount you stand to win and the probability of winning. Fees are smaller nearer to 0¢ and 100¢ than near 50¢, but that’s because a) you stand to win a lot less when buying near 100¢ and b) you win less often when buying near 0¢. The end result is that that your long-term commission will be equivalent to a rate of 1.75% for makers and 7% for takers.

To confirm this is correct, I calculated and compared the two methods using R software at three different price levels (see below). Note that this assumes each bet’s probability of winning is in line with the price. If you are a profitable bettor then your bets will win more often, meaning your fees paid will end up a bit lower than the equivalent Betfair commission rate, and vice versa for long-term losers.

So how bad is 7% commission? If a market is a coin flip (e.g. an NFL or NBA handicap market) then buying at 50¢ on Kalshi would be buying at 51.75¢ after the taker fee. This is equivalent to 1.932 decimal or -107 in American odds. It’s not amazing value, but it’s good enough that a recreational punter will accept it for the convenience of a single app where they can bet on seemingly anything — especially if they live in a large state like California or Texas where there are few other legal options.

The proposition for makers is much more favourable, as the 1.75% commission makes the same bet equivalent to 1.983 decimal or -101.8 in American odds.

To illustrate the long-term effect of commission, the below chart shows the profit distribution after 1,000 bets for a punter who has an edge: he is consistently buying bets at 50¢ that have a 52% chance of winning. This may not sound like much of an edge but this would put you in elite territory in any high-liquidity market!

In the fantasy scenario of zero fees, the expected profit after 1,000 bets is +$4,000 with an 89.1% probability of being in profit. At 1.75% the expected profit drops to +$3,125, and Betfair’s 2% commission eats up more than a quarter of your long-term profits as the expected +$4,000 drops to $2,960. But on Kalshi’s 7% fees your expected profit would be only +$500 with a 56.3% chance of being ahead and 43.7% probability of being down overall.

So this begs the question: why ever be a taker?

1. Recreational punters. Some people just aren’t price sensitive, or they can’t be bothered to waste time shopping around for the best price. If I’m being honest, this is my approach to lots of things in life! It would be exhausting to try and secure the best price every time I pay for something, so I totally understand this behaviour.

2. You have a big edge thanks to some new information. If you found out about some major event (e.g. a goal or a big injury) more quickly than anyone else, then obviously you are happy to pay 7% fees if you know that the market is dramatically wrong and the price is likely to change imminently. Fill your boots!

3. You have a special rebate deal (more on that shortly).

In the absence of the above, it seems like anyone with the goal of making a long-term profit on Kalshi is forced into being a market-maker. The disadvantage for makers is that you may not get matched, or you may only get matched because of some new piece of information (see 2. above). Because of these factors, queue position becomes very important: if you are at the back of a very long queue then it becomes less likely that you get matched in the normal course of events, whereas if a goal or major injury occurs then you are absolutely definitely getting matched straight away, and you won’t be happy about it.

There’s another major difference between Kalshi and Betfair’s commission that has a huge effect on strategy. Kalshi charges a fee on every single bet, whereas Betfair only charges commission on your net position when the market is settled. This makes trading or micro-trading strategies a lot more viable on Betfair than on Kalshi.

For example, suppose you buy 10,000 contracts at 50¢ and sell them back at 53¢. Congratulations, you just made $300! On Betfair you would pay 2% commission on this $300, so your net win is $294. On Kalshi you would pay fees on both the buy AND the sell, so for makers the net win would be $300 - $43.75 - $43.59 = $212.66 while for takers it would be $300 - $175 - $174.37 = a net LOSS of $49.37!

Although on the face of it Betfair are leaving a lot of money on the table by only charging commission on net winnings, it does have a couple of advantages. Firstly, a trading strategy allows you to place much larger stakes as there is less of your bank at risk, assuming you can trade out if the price goes against you (this is a BIG assumption). And secondly, when you execute your second trade there is another party to that bet, so you are recycling money back into the market. It’s like the multiplier effect in economics. So maybe this approach is good for the ecosystem? I don’t have the answer to that, but I suspect it’s something that will vary depending on the liquidity and volatility of each particular market.

Finally, let’s address the elephant in the room. There are big players operating in the market with a huge advantage: rebate deals. In exchange for adding large volume, some organizations sign Market Maker deals which effectively give them much lower fees per bet than the average punter. This obviously puts them in a strong position as they may be able to execute ‘taker’ trades without the prohibitive costs. I won’t pretend to know all the inner workings of these deals, and I should also point out that Kalshi has other incentive programs to encourage liquidity.

Originally this was just some private research as I was considering whether Kalshi is worth my time & effort to see if it could be a viable income source. I realised this might be of interest to others, so I turned it into this article. So, is Kalshi worth it? I’m going to say yes overall, simply because of the amount of money sloshing around from price-insensitive, inexperienced Americans. And there’s never any harm in keeping your tools sharp of course; I’ve been out of the game for a few years but it’s always good to be ready for the next opportunity.

What about the Kalshi ecosystem? Is that amenable to long-term health? That’s not my area of expertise, but I’m sure Kalshi recruited people with industry experience and came up with this very deliberate two-tiered fee system. There’s obviously a first-mover advantage with these things and everyone is familiar with Kalshi now, so it feels like they will have to either mess it up or get banned by the government, otherwise the ceiling is very high. I’ve only had a few bets so far — fortunately (aftertiming alert) one of those was on Spain to win the World Cup — but I’m going to continue having a play around with it and see what happens. Small stakes to begin with, of course. Please bet responsibly! And if anyone with more experience of Kalshi has anything to add, I would love to hear it.

In part one of my 2025-26 Premier League season review I looked at managers – particularly teams who had changed manager and how it had affected their performances. After my World Cup hiatus I will now present part two, focusing on players.

It’s not easy to isolate individual player contributions in a fluid team sport like football because many actions may be accumulated at the expense of team play. For example, an attacker may increase his goal involvements by staying forward instead of helping out in defence, or a defender may win more tackles and interceptions by constantly wandering out of position in search of the ball.

A more simple approach is what I like to call proof is in the pudding: does the team play better or worse when Player X is in the team? I used my contexG ratings for each match and looked at the difference when each player starts or does not start.

Now, I know a lot of people just want to scroll down to the results, so I’ll show them first and explain the methodology later.

Players associated with POSITIVE team effects

The #1 player was Leeds United’s Pascal Struijk with an estimated difference of +0.29 goals per game. When I first produced these results at the start of the summer this was unremarkable; it became a lot more interesting when Struijk was purchased by Tony Bloom’s shrewd Brighton operation for somewhere between £15m-20m. Struijk started 33 games for Leeds last season and among the five he missed were Leeds’ three worst performances of the season according to contexG. Coincidentally, two of these were against Brighton: the 3-0 away defeat in November and the highly fortuitous 1-0 win in the penultimate game of the season.

Brighton’s top player in the model was Pascal Gross (+0.18 goals). Following Gross’s arrival during the January transfer window, Brighton saw a clear uptick in form and they look strong going into next season. They sold key defender Jan Paul Van Hecke to Spurs in the summer but with the arrival of Struijk along with top Spurs prospect Luka Vuskovic, it would be no surprise if the Seagulls come out smelling of roses.

Second on the list is Everton midfielder Kiernan Dewsbury-Hall. He was Leicester’s player of the season when winning the Championship before becoming a forgotten man after he joined Chelsea — certainly not the first player to suffer that fate. Dewsbury-Hall had a very solid first season at Everton after an eyebrow-raising £29m transfer, contributing 8 goals and 4 assists. A note of caution is required concerning his +0.29 contexG effect, however, because Dewsbury-Hall’s stretch of missed games in December and January almost perfectly overlapped with the injury absence of Iliman Ndiaye (+0.21).

Similar caveats apply to the positive effects of Crystal Palace players Daichi Kamada, Adam Wharton and Ismaila Sarr. Thanks to Palace’s Conference League run, Oliver Glasner was quite often resting his best players in the same matches, so it’s hard to isolate which of them is the true star (if any) and which are riding on the coattails. But I think it’s pretty clear which of the three would currently fetch £100m in the transfer market.

Newcastle fans will be pleased that despite the loss of Anthony Gordon and Sandro Tonali to Barcelona & Spurs, they have (so far) held onto Bruno Guimaraes who had a +0.17 contexG effect. With Alexander Isak leaving a year prior, one can’t imagine Bruno is thrilled at the situation. He’s been linked with Arsenal who made a derisory offer earlier in the summer, but if a club can stomach a fee nearer to the Tonali one for a player who turns 29 in November, it’s not hard to see a deal being struck.

Players associated with NEGATIVE team effects

The biggest negative effect in the league belonged to Joshua Zirkzee (-0.33 goals), a player who feels emblematic of Manchester United’s woeful recruitment prior to the Ratcliffe takeover of football operations. Zirkzee started in some of United’s worst performances of 2025-26: the 0-1 defeat at home to 10-man Everton; the 1-1 draw at home to West Ham; the 1-1 draw at home to Wolves; and the 0-0 draw at Sunderland. It would be a surprise if he remains at the club for a third season.

Second in the negative chart is Brajan Gruda (-0.33) who started in a series of disappointing performances for Brighton in the first half of the season before departing on loan to RB Leipzig in January. As noted earlier, Brighton’s performances improved markedly after the midseason transfer window which doesn’t do Gruda’s reputation any favours. He has returned to Leipzig for another loan in 2026-27 with an obligation to buy if conditions are met.

Liverpool fans will be concerned, although probably not surprised, at the prominence of Alexander Isak with a -0.20 goal effect. Isak only managed 8 league starts for his new club, including defeats in each of the first four. There were mitigating circumstances for Isak as he missed preseason due to the protracted and acrimonious transfer from Newcastle, followed by a broken leg sustained in the act of scoring at Spurs, but a serious improvement will be required next season from a player who cost a whopping £125m.

Much like in the positive player table, multiple Crystal Palace players (Will Hughes, Justin Devenny and Brennan Johnson) appear in the negative table and similar caveats apply as these backups were often fielded together in low-priority matches before key European ties.

Best and worst effect for each team

There are a few notable names in the above chart which shows the best and worst player for each team in terms of estimated contexG effect. Leandro Trossard led the way for Arsenal but has been sold to Besiktas. Meanwhile, Kai Havertz started 7 games after returning from injury, coinciding with a drop-off in Arsenal’s performances — albeit still accumulating enough narrow wins to secure the title.

Donyell Malen was on fire for Roma after moving from Aston Villa in January, scoring 14 goals in 18 matches, but he had Villa’s most negative effect (-0.20) while they generally played their best football with Pau Torres (+0.16) in the lineup.

Liverpool’s season was disappointing but Huge Ekitike (+0.20) was one of the bright spots and the Reds’ season really tailed off after the Achilles injury that will keep him out for a good chunk of next season.

World Cup winner Pedro Porro led Spurs in chances created last season — admittedly not a high bar, but his ball progression and creativity from the right-back position makes up for some suspect defending, and Spurs will be pleased to have tied him down until 2031 amid interest from Manchester City.

Speaking of City, their most negative impact belonged to John Stones, now a free agent after being released at the end of the season. Stones’ presence in the England squad ahead of the likes of Harry Maguire was surprising to many, and according to reports he is now close to signing for Inter.

Max Kilman had the biggest negative effect for West Ham (-0.19) and he was one of those buys that looked bad at the time and looks even worse now. The £40m signing made the last of his 17 starts in the shambolic defeat at his former club Wolves when the Hammers were 3-0 down at half time. On the flip side, Taty Castellanos (+0.22) impressed after joining in the January transfer window, and the sales of Mateus Fernandes and Crysencio Summerville for combined fees of more than £130m open the possibility of keeping Castellanos alongside the likes of Jarrod Bowen in a squad that should be far too strong for the Championship.

Methodology

As mentioned earlier, these results are based on team-level performance using my contexG model which differs slightly from traditional xG:

• ContexG accounts for game state (goals, red cards).

• The xG for any single shot is capped, because the creation of high-xG chances involves an inherent amount of luck (e.g. goalkeeper spills a long-range shot leaving an open goal).

• ContexG is trained on future team performance, not on whether the shot itself is scored.

For each player, I compared the team’s contexG (adjusted for opponent and home advantage) in games where they started versus games where they didn’t start.

It’s important to note that some players are easier to compare than others, because you need a large sample of games with the player and a large sample without the player in order to make a meaningful comparison. An even split of 19 games played and 19 games missed would be ideal; a player starting all 38 games (or missing them all!) would contain no statistical power whatsoever.

The raw on-off differences were multiplied by an empirical Bayes shrinkage factor to get the final results above. Clearly, there are many factors that can affect a team’s performance from one game to the next, and any association with an individual player should be treated cautiously and combined with your prior belief about the player’s value. A potentially more accurate model of player value could be built using a more granular minute-by-minute on-off comparison, but this is also prone to overfitting so I would again urge caution and common sense.

Finally, I just wanted to mention that I had planned to make a lot more content regarding the World Cup but due to a number of family & work commitments I eventually just decided to have a bit of time off from writing & modeling and watched the tournament as a fan. There was plenty of great coverage from Nate Silver, Michael Caley & others, and I don’t think World Cup summers will ever be a time where we are lacking good content. I’m feeling recharged and ready to write some new articles as the domestic season approaches.

I’m starting my review of the 2025-26 Premier League season by looking at managers. Specifically, teams that changed managers either before or during the season, and the associated change in performance or style.

One of the performance measures you will see is contexG. For those who are not familiar with my contexG model, it is similar to expected goals (xG) but takes into account game state (score, time elapsed, red cards) and certain match stats, and it is trained on future team performance rather than the goals scored in that particular game.

The value of such a model can be illustrated by looking at the closing table and contexG ratings from the 2024-25 season. The four worst teams who stayed up, according to contexG, were Tottenham, West Ham, Wolves and Nottingham Forest—all of whom became the main relegation protagonists in 2025-26.

Followers of the model would also have been unsurprised by the high finishes for Brighton, Bournemouth and Brentford this season, as well as the resurgence of Manchester United. Speaking of whom, that’s where I will begin this review:

Manchester United: Compared to Amorim, Carrick was more defensive, more effective…more fortunate?

As the above table demonstrates, United had finished the previous season more strongly than their 15th-place finish would indicate. Unfortunate to lose the Europa League final to Spurs after outshooting them 16-3, United returned home and took out their frustration on Aston Villa in a surprisingly one-sided home win (even before the Martinez red card) to deny in-form Villa a Champions League spot.

Sesko, Mbeumo, Cunha and Lammens were then signed in the summer for more than £200 million and, with no European football, expectations for the league season were high. Performances were reasonable under Ruben Amorim but, despite big edges in shots and xG, he could only manage 1.6 goals per game for and against before being sacked after 20 games. The sacking was typically well-timed by the club’s board, coming just as Bryan Mbeumo, Amad Diallo and Noussair Mazraoui returned from AFCON and Bruno Fernandes returned from injury.

Results under Michael Carrick were excellent with United getting 2.29 points per game (87 points over a full season). But the underlying stats were not drastically different: contexGD increased modestly from +0.30 to +0.39 and xGD from +0.53 to +0.6.

The obvious change Carrick made was a shift from Amorim’s 3-4-3, with Bruno Fernandes playing centre-mid, to a 4-2-3-1 with Fernandes in a more advanced role and Kobbie Mainoo forming a double pivot alongside Casemiro. The stats show that this led to a more defensive approach: total xG in games under Carrick was 2.78 compared to 3.35 under Amorim while United’s shots, possession and field tilt all decreased.

The parallels between Carrick and Ole Gunnar Solskjaer are obvious and it is perhaps a surprise that United have repeated the same pattern of promoting an under-qualified interim to a permanent role after a short period of success. Massively overperforming your numbers, plus having a load of extra Champions League fixtures after a Europe-free season, is a recipe for potential regression. But sometimes football management is about timing, and if United can recruit well this summer (presumably they will spend another £200m or so) while avoiding injuries next season then maybe Carrick will preside over a new era of success. Overall though, it’s not hard to see this ending badly.

Tottenham Hotspur: From bad, to worse, to just about good enough

Spurs were one of the more intriguing teams coming into the 2025-26 season. After a decent start in 2024-25, a huge injury crisis plus a focus on the Europa League saw them plummet like a stone and finish in 17th place. However, the fact their goal difference was still positive going into the final day is an indication that this was not like your ordinary 17th-placed team. The above 2024-25 table shows their contexGD over the season was -0.02, but when weighted towards recent results it was -0.54.

So which version of Spurs were we going to see in 2025-26 under Thomas Frank? It felt like the possible range of outcomes was wider than most, but the reality under Frank was dismal, bottom-half football. 1.12 points per game (42.6 points over a full season), -0.18 contexGD and near the bottom of the league in shots and xG. In fact, Spurs’ results were artificially held up for a time thanks to positive finishing luck and Frank’s trademark set piece prowess—surprisingly, they ended the season with as many goals scored from corners as Arsenal.

Another huge injury crisis didn’t help matters, but it had become pretty obvious Thomas Frank was not the right man for a club that is associated with attractive, passing football. Igor Tudor came in, ostensibly because of his prior experience with similar “firefighting” assignments in Italy. Tudor was an unexpected and uninspiring appointment but still, it couldn’t get any worse could it?

Well actually, yes it could, and by a huge margin. In the five games under Tudor, Spurs were literally the worst team in the Premier League, gaining a solitary point and averaging -1.07 contexGD—worse than Burnley over the full season. At a time where simplicity seemed like a good idea, Tudor employed weird formations with players out of position: the Fulham away game was a good example as Xavi Simons was played at left-midfield and Conor Gallagher right-midfield in a 4-4-2, with both players completely ineffective and subsequently dropped by Tudor.

Relegation had now become a very real possibility which meant Tudor was swiftly shown the exit, and Roberto De Zerbi was appointed for the final 7 matches. Early signs were not good as his opening defeat at Sunderland largely followed the Tudor performance trend. But interestingly, having started youngsters Archie Gray and Lucas Bergvall in midfield in that game, neither started again under De Zerbi who opted for the more seasoned Bentancur, Gallagher and Palhinha, as well as Xavi Simons prior to his ACL injury.

Following that shift towards experience, Spurs were actually pretty good to finish the season. Their contexGD under De Zerbi was +0.11 and they were now back to dominating games (62.4% field tilt) largely thanks to a huge improvement defensively, conceding just 7 goals from 7.49 xG across the 7 games.

Of course, part of that defensive improvement could be attributed to the desperation of a relegation fight, and there is no guarantee it will continue into next season. But signs are certainly more positive under De Zerbi than the absolute shambles he inherited.

West Ham: They just couldn’t defend

The general narrative of West Ham this season is that Nuno took over a mess from Graham Potter and eventually turned it around but it was too late to avoid the drop. The season-long contexGD chart below would support that narrative as the first half of the season is dominated by large red bars before small green shoots emerged after the new year.

It is no surprise that Nuno switched to a more counterattacking style relative to Graham Potter with possession dropping from 50.2% to 44.0% and field tilt down to just 40.7%. However, this change did not elicit a defensive improvement: in fact, both xG for and xG against increased dramatically under Nuno as the total xG in West Ham’s matches rose from 2.64 to 3.34.

I think this is why it seemed to many observers that West Ham were playing well on the “eye test” which is generally biased towards attacking play. In a game with many chances it seems that both teams have played well, and in a tight 0-0 we say both teams were poor, but that logic isn’t really justified—there have been plenty of dull finals between the very best teams in major tournaments.

The notable improvement after the new year could be associated with a couple of personnel changes. Taty Castellanos arrived from Lazio and his all-around game seemed to immediately transform West Ham’s attack. Meanwhile, cynics might point to the departure of the much-maligned Lucas Paqueta as also giving the squad a boost.

West Ham put themselves in position to escape but their final two away defeats seem emblematic: 3-0 at Brentford and 3-1 at Newcastle; with everything on the line they just got shredded open too easily.

The club have announced that Nuno will be staying as manager, which feels like a plus, and despite the likelihood of losing Bowen, Summerville, Fernandes and Castellanos it still feels like West Ham should be able to bounce back at the first attempt. They finished the season with a recency-weighted contexGD of -0.41 which is the highest since Fulham were relegated in 2020-21 with -0.22. Encouragingly for West Ham, the Cottagers then won the Championship with 90 points and a +63 goal difference.

Chelsea: Pretty good until the players completely gave up on Rosenior

It’s hard setting expectations for Chelsea. They are among the biggest spenders in world football and yet under the new ownership they have gone 12th, 6th, 4th, 10th.

Enzo Maresca had them in reasonable shape: a contexGD of +0.52 is not too far from a title challenge conversation, and a field tilt of 59.3% is much more in line with a team with title aspirations compared to, say, Carrick’s United.

Maresca was sacked for apparently being in discussions to take over from Pep Guardiola at Manchester City, which tells you that a) City, who are fairly well-run, rate him highly, and b) Maresca was happy to jump ship from Chelsea despite the abundance of young talent and money sloshing about.

After the questionable appointment of Liam Rosenior as Maresca’s replacement, performances didn’t suffer too badly at first. Rosenior experienced two unlikely home draws thanks to Leeds scoring twice from 4 shots and then a Fofana red card (a theme of the season) throwing away two points against Burnley. This was followed by a 4-1 away thrashing of Aston Villa at which point a Champions League spot looked probable.

But then the wheels came off as Chelsea lost their next five matches, culminating in a pathetic defeat at Brighton which made clear the players had given up on the coach. Following Rosenior’s sacking, Cole Palmer posted an Instagram meme implying the players were now “free” from the apparently unpopular boss.

Rosenior’s stint actually doesn’t look too bad overall, with a fair amount of bad luck including the aforementioned home draws plus a defeat to Manchester United after outshooting them 21 to 4. Unsurprisingly, after downing tools for Rosenior the players didn’t pick them back up again for interim manager Calum McFarlane and the season tailed off badly, resulting in a 10th-place finish. Xabi Alonso comes in to tidy up the mess, and it feels like there is plenty of potential to rally this talented squad to an immediate high finish next season as long as the financial situation isn’t too dire.

Nottingham Forest: A mad season ended in safety thanks to prolific finishing down the stretch

A wild season for Nottingham Forest saw four managers, a relegation battle and a European semi-final. Despite flirting with a Champions League finish the previous season, Nuno Espirito Santo was sacked just three games into the season after his relationship with owner Evangelos Marinakis deteriorated. That was followed by the short but disastrous reign of Ange Postecoglou who managed just one point and one goal from his five league matches in charge before also being given the boot.

Sean Dyche came in to steady the ship and did a reasonable job before being sacked after the 0-0 draw at home to Wolves. Ironically, as you can see from the large green bar in the chart below, contexG rated that as Forest’s best performance of the entire season. It was an absolute battering: 35 shots, 40 touches in the box, 3.69 xG on Understat…you could hardly blame the manager for those chances not going in.

But Marinakis is unpredictable, so he replaced Dyche with Vitor Pereira who benefitted from a series of matches in which Forest were prolific finishers: 2 goals from 9 shots at Manchester City, 3 goals from 8 shots at Spurs, 4 goals from 10 shots versus Burnley, 5 goals from 10 shots at Sunderland, and 3 goals from 6 shots at Chelsea. In those five games alone, Forest scored 17 goals from 43 shots, whereas they had scored 0 from 35 in Dyche’s finale. As Jimmy Greaves would say, it’s a funny old game.

The overall stats don’t really show a huge uptick under Pereira. His opponent-adjusted contexG was -0.45 which is the worst of any of the teams who stayed up. Expected goal difference is even worse (-1.47) with possession, field tilt, progressive passes and crosses all the lowest of Forest’s four managers.

It’s hard to imagine Forest’s financial situation is especially healthy. Having already encountered previous PSR issues, last summer they spent north of £200m with questionable return. Star midfielder Elliot Anderson sounds like he is off to Manchester City and it feels like the outlook for Forest is gloomy. Like other clubs, they will be hoping the gulf between the Premier League and Championship means they could be pretty bad next season but not bad enough to go down as long as none of the promoted clubs “do a Sunderland”.

Brentford: Andrews replaces Frank, business as usual

Comparing the stats from Thomas Frank in 2024-25 to Keith Andrews in 2025-26, it looks like Brentford just sort of rumbled on doing the same thing under a different coach. They are clearly a well-run club and anyone thinking they will steal Brentford’s edge just by appointing their manager will be sorely disappointed. To be honest, I think if Dean Smith was still in charge for all these years, the Bees would probably be in a similar position to where they are now.

The style of play is familiar: Brentford play a passive game, allowing opponents to have more shots than themselves but excelling at high-xg fast breaks and set pieces. After a flirtation with the Champions League race, a disappointing finish meant Brentford actually ended up with fewer points (53) than the previous season (56). Andrews wasn’t helped by quite a big finishing underperformance (1.47 goals scored per game from 1.78 xG), and overall it’s a creditable effort after losing Mbeumo & Wissa and starting the season among the relegation favourites. Missing out on Europe was a blow, but it means they will get a clear run at the league next season. Another comfy mid-table finish seems likely, with some upside if they get better dice rolls.

Wolves: Bad but unlucky under Pereira, then just bad under Edwards

It was an odd start to the season for Wolves as they were bad, but not 2 points from 11 games bad. Nonetheless, out went Vitor Pereira and former Wolves player Rob Edwards arrived, adding some performative energy on the touchline but ultimately not improving the team. Under Edwards, Wolves retreated into a more defensive shape, with field tilt dipping below 40%. Even against woeful Burnley in the final game, they played on the break with 30% possession. Grim stuff.

Thumbs up, thumbs down

So which changes worked and which didn’t? First the good ones. It’s hard to argue that Carrick wasn’t good for United, at least in the short-term. The question is whether it was a good idea to keep him permanently, which I guess we will find out next year. Roberto De Zerbi—an appointment many criticised at the time—did an excellent job at Spurs and looks like a good fit to lead them into next season.

On the flipside, Thomas Frank was never a good fit for Tottenham Hotspur, but he looks like Bill Nicholson in comparison to the hapless Igor Tudor who almost got the club relegated. Liam Rosenior, formerly of Strasbourg and Hull City, was an appointment that every man and his uncle knew would turn out poorly at Chelsea. Likewise, nobody expected Ange Postecoglou to come into Forest mid-season and successfully introduce a polar opposite playing style.

The rest of the changes are much of a muchness. I often wonder: what if football teams didn’t have managers? Teams would still experience dramatic fluctuations due to good players arriving or leaving, dressing room harmony, injuries, luck, and other random factors. But there is a tendency to attribute all the success or failure of a football club to the manager, echoing the human comfort found in the Great Man Theory. We like simple explanations for phenomena rather than putting things down to fluke, or a combination of smaller factors, or simply “I don’t know”.

A paradox exists where good football managers seem like they are incredibly valuable, and yet simultaneously it is very hard to identify who is a good manager. The head coach from Team A does a brilliant job and is poached by the richer Team B, only to then look useless. I don’t need to give you an example because I’m sure you can think of 20 examples pretty easily.

Even Pep Guardiola still gets the “but could he do it at a smaller club?” critique, which is fair enough because the challenge of coaching the world’s best players is very different to setting up a team to defend against superior opposition. And perhaps therein lies the answer: there aren’t necessarily good managers and bad managers, just the right manager for the right club at the right time.

In my next review, I will take a look at which players have been associated with an increase or decrease in their teams’ performance levels. It’s a difficult area to find genuine effects, but it should still be interesting to explore.

On Wednesday night, Aston Villa face Freiburg in the 2026 Europa League final. This follows last year’s final where Tottenham (17th in the Premier League) beat Manchester United (16th), prompting many to ask whether the Europa League has been weakened since the revised format.

Below is a plot comparing the Club Elo ranking for the finalists of every UEFA Cup and Europa League iteration, with Freiburg currently ranked 48th in Europe. How far back do you have to go to find another finalist ranked as low as 48th? In fact you only have to go back four years to find Rangers, who lost on penalties to Eintracht Frankfurt in 2022.

The Rangers-Frankfurt final followed a relatively strong decade of Europa League finals from 2012-2021 where 90% of teams were ranked in the top 20 of the Elo ratings. A notable exception was Dnipro Dnipetrovsk (55th) in 2015 who scraped through the group stage despite losing 3 of 6 games, then surprisingly reached the final before losing 3-2 to Unai Emery’s Sevilla. Incidentally, this was also the first season where the winners were rewarded with a place in the following season’s Champions League.

From 1999 to 2024, the strength of the UEFA Cup (rebranded as the Europa League in 2009) was boosted by the addition of teams that finished 3rd in the group stages of the Champions League. Across these 25 seasons, 19 Europa League finalists had dropped down into the competition via this method, with 9 of them lifting the trophy.

Prior to 1997 the European Cup was literally just for the champions of each country, plus the previous year’s holders. This left room for many elite teams to be playing in the UEFA Cup, and it was not unusual for the team ranked #1 in Europe to be playing in the UEFA Cup final rather than the European Cup final. The thin divide between the two competitions is well illustrated by Borussia Monchengladbach (1975) and Liverpool (1976) who both won the UEFA Cup as the #1-ranked team in Europe, before playing against one another in the 1977 European Cup final, which Liverpool won 3-1.

One occasion when the top-ranked side did not win the UEFA Cup final was in 1982 when Sven-Goran Eriksson’s IFK Gothenburg side (ranked 64th) caused a big upset over that season’s Bundesliga champions Hamburg. Meanwhile, the lowest ever ranked team to make a UEFA Cup final were Hungary’s Videoton (91st) who lost to Real Madrid in the 1985 final, having knocked out Manchester United along the way.

Since the expansion of the Champions League in 1997 to include non-champions from the strongest domestic leagues, it has become far less likely that the #1 ranked team in Europe could make the final of the UEFA Cup / Europa League. In fact, the only time this has happened was in 2004 when Valencia romped to the La Liga title under Rafael Benitez before beating Marseille 2-0 in the UEFA Cup final. Since then, no finalists have ranked higher than 4th in the Elo ratings.

When we look at the same plot for Champions League finalists, the difference between the European Cup era and the Champions League era is stark. When it was a straight knockout featuring only the champions of each country, there was a chance that rank outsiders such as Panathinaikos, Malmo or Partizan Belgrade could get to a final. But there currently haven’t been any teams from outside the top 10 to reach the final since Liverpool in 2005.

That’s partly a self-fulfilling prophecy, of course, because the strength and format of the modern competition dictates that any team reaching the final will have picked up a lot of Elo points by virtue of their results along the way! Nonetheless, it is hard to argue against the fact that Europe’s showpiece final is now more likely to feature two of its best teams.

For more context, we can plot all the UEFA finals on one chart, including the newly-introduced Conference League and the defunct Cup Winners’ Cup.

Panathinaikos are rated as the weakest team ever to play in a UEFA final, ranked a lowly 214th in 1971. At the time, Greece really had no European pedigree—prior to that season, Greek teams had got past the first round only once in 12 attempts in the European Cup, and only twice in the Cup Winners’ Cup where Olympiacos had lost in the second round to West Ham and Dunfermline. Panathinaikos reached the final thanks to away-goal victories over Everton and Red Star, before losing 2-0 to Ajax at Wembley.

Since the introduction of the Conference League in 2021-22 it would appear that the quality of Europa League finals declined—even when third-placed teams were still dropping down from the Champions League. This suggests UEFA may have turned one stronger competition into two weaker ones, but it is still a small sample.

So to revisit the original question: have Europa League finals become weaker? Possibly a little bit compared to very recent history, but I think we also tend to forget some of the less glamorous finals from the past. Rangers v Eintracht Frankfurt and Sevilla v Dnipro were two examples but there was also Atletico Madrid v Fulham (2010) and Rangers v Zenit St. Petersburg (2008), while Feyenoord (ranked 54th) upset Borussia Dortmund in 2002.

It’s important to add that I have only looked at the finalists here. When Eintracht Frankfurt reached the Europa League final, for example, they had knocked out Barcelona along the way. In contrast, this year Aston Villa have played Lille, Bologna and Nottingham Forest—and that was the difficult half of the draw.

I plan to develop this analysis further in a future article where I will take a more mathematical approach and compare the probabilities of winning the whole competition. An important aspect of this is the difficulty of getting into the competition in the first place! Strong clubs in weak leagues, such as Celtic and Rangers, have benefitted from being in Europe basically every season. On the other hand, the champions of England or Germany had a great chance of winning the old European Cup once they were in it, but it was very hard to win the league in the first place.

There will always be the debate between the old European Cup and the modern Champions League, or the old UEFA Cup and the modern Europa League. It’s impossible to say that one is objectively harder than the other—there is only one champion each season, after all—but the probabilities will shift based on which perspective you are considering. And when it comes to football, it’s funny how supporters’ views will often align with the one that puts their own team in a good light.

But that’s why shocks are called shocks! They don’t tend to be expected. So I was wondering, if we look back at the history of the English top flight, which teams historically have had the highest probability of winning the league only to fail at the end? And how would we quantify that?

I created a model that could give the probability on any date of each team becoming champions. Here are the basic details:

• Daily team ratings were taken from Club Elo.

• An adjusted negative binomial model was created to generate score probabilities.

• Club Elo ratings begin in 1940, so only seasons since the Second World War were analysed (sorry, Huddersfield fans).

• At the end of every match day, standings were calculated and the remainder of the season was simulated 10,000 times.

• Only dates with 6 or fewer remaining matches were included; this is about late collapses rather than well-fancied teams having a bad year (it also saves a lot of time running the simulations). So those of you expecting Kevin Keegan to feature here will be disappointed.

• League tables were calculated according to the correct rules for the time—so that’s goal average up to 1976, and 2 points for a win until 1981.

Without further ado, here are the top five teams who had a great chance to be champions but couldn’t do the business:

5. Burnley 1961-1962 (Probability: 80.4%)

If you had asked me to name every club to have been champions of England, I’ll be honest, Ipswich Town might be one of the trickier ones. But when I scratched beneath the surface their 1961-1962 win emerges as one of the more fascinating and unexpected championships in football league history.

At the time, Burnley were a true force in English football, having won the league in 1960 and in the midst of a run of 12 consecutive top half finishes. The dominant team of the era were Bill Nicholson’s Spurs, coming off a historic league & cup double the previous season, and this year reaching the European Cup semi-final where they would lose to Benfica.

There was also an emerging Everton side who were about to embark on a 10-year stretch where they won two league titles and finished outside the top 6 only once.

Ipswich, meanwhile, were playing in the top flight for the first time in the club’s history under manager Alf Ramsey, a former Spurs right back who learned his trade under Arthur Rowe in Tottenham’s influential “push and run” champions of 1951.

Despite trailing by a point, on the 14th of April Burnley had two games in hand and as the strongest team were still considered 80% favourites, with the next two games at home to Lancashire rivals Blackburn and Blackpool.

One thing this model doesn’t account for is fixture congestion. Burnley had played an FA Cup semi-final replay on the 9th of April, and faced a staggering 8 further league games in the remainder of the month, including three games in four days.

Burnley’s finish was grim: just one win from their final eight games. Ipswich became 61% favourites after a 3-0 win at Arsenal in their penultimate match, before sealing the championship with a 2-0 victory over Aston Villa with star striker Ray Crawford scoring both goals.

Burnley still had a game remaining after the title race was over, which they duly lost 4-0 to Sheffield Wednesday, and then to rub salt into the wounds they were beaten by Spurs in the FA Cup Final.

For Ipswich, it really seems like they caught lightning in a bottle for that one season. They were thrashed 5-1 by Spurs in the following year’s Charity Shield and lost to AC Milan in the first round of the European Cup.

By October 1962, Ramsey had already agreed to take over the England job the following summer, leading them to a World Cup triumph in 1966. Ipswich would end the season in 17th place and then were relegated rock bottom a year later.

But this was far from the end of the good times in Suffolk – a decade later, another future England manager named Bobby Robson would oversee a more sustained period of success including an FA Cup, UEFA Cup and league finishes of 4th, 4th, 3rd, 3rd, 3rd, 2nd, 2nd…but not 1st.

4. Manchester United 1991-1992 (Probability: 80.9%)

The 1991-1992 season is often seen as the final year of the old era due to the impending rebrand of Football League Division One to the Premier League. But it was also the final year of Manchester United’s long league title drought since the days of Charlton, Best, and Law in 1967.

With the benefit of hindsight, of course, United’s late collapse in 1992 is a mere footnote beneath the outrageous success of the Alex Ferguson era. But what a collapse it was!

As the previous year’s champions Arsenal had mounted a disappointing defence of their title and perennial powerhouses Liverpool declined under the disastrous reign of Graeme Souness, the Red Devils found themselves in a commanding position on the 16th of April, two points clear of Leeds United with a game in hand.

Leeds had only been promoted two years beforehand but had finished a creditable 4th in the previous campaign, and boasted one of those great midfields that is still easy to rattle off today: Speed, McAllister, Batty, Strachan. They also signed a chap called Eric Cantona from Marseille in January after he was turned down by Souness at Liverpool and Sheffield Wednesday couldn’t afford his wages.

The impact of Cantona at Leeds that season is perhaps overstated (he only made 6 starts, scoring twice) but it remains a remarkable fact that Cantona won the title in every season he played in England, other than 1994-1995 when he was suspended from January onwards after attacking a supporter at Selhurst Park.

With five games to go, the model gave Manchester United an 81% chance of finally ending their barren run, and even after a 1-1 draw at Luton things were still looking good.

And then the wheels truly came off.

In a grueling run of fixtures, Fergie’s men contested the League Cup Final on 12 April, defeating Nottingham Forest 1-0, and then played league matches on the 16th, 18th, 20th, 22nd, and 26th.

It was Forest who inflicted the first major defeat, getting revenge for Wembley with a 2-1 win at Old Trafford that sent the title race into coin flip territory. And then, shockingly, United lost 1-0 to a West Ham team that had lost 10 of its last 13 games and would be relegated in bottom place. All of a sudden, Leeds had a 75% probability of winning the league.

To cap it off, it was hated rivals Liverpool who dealt the fatal blow with a 2-0 win at Anfield that saw Ian Rush finally score his first goal against Manchester United at the 24th attempt!

Having won 3-2 at Sheffield United earlier in the day, Leeds were crowned champions on the sofa—literally in this case, as ITV staged a watch party with Cantona, Batty and McAllister in Lee Chapman’s living room. And 33 years later, Howard Wilkinson still remains the last Englishman to manage a title-winning side in England.

3. Liverpool 2013-2014 (Probability: 82.2%)

Although Liverpool have since become regular contenders for the Premier League, the 2014 title challenge came a little bit out of the blue. Still recovering from the disastrous Hicks-Gillett ownership era, the Reds had only finished 7th, 6th, 8th, and 7th in the previous four campaigns.

But Liverpool had a few things that clicked together this year. Thanks to a lack of European football and early domestic cup exits away to Arsenal and Man Utd, Brendan Rodgers’ side only played 43 matches all season compared to 57 each for Man City and Chelsea.

And up front, the emergence of teenager Raheem Sterling alongside Daniel Sturridge and the genuinely world-class performances of Luis Suarez electrified a forward line that had until recently included £35m flop Andy Carroll.

Following the epic 3-2 win over Man City and an away victory at Norwich by the same scoreline, Liverpool found themselves five points clear of Chelsea and nine clear of City. The model gave them an 82% probability of lifting a first league title since 1990.

It would require seven points from the final three games to guarantee the championship, and Liverpool’s prospects were further improved by the fact that Chelsea, with only slim odds of the title themselves and preparing for a crucial Champions League semi-final second leg against Atletico Madrid, decided to rest several key players for the Anfield clash on 27 April (ironically, one of the backups who played as a result was a certain Mohamed Salah).

Alas, a Mourinho defensive masterclass ensued, and the surprising 0-2 defeat to Chelsea, including the now infamous slip by Steven Gerrard to gift Demba Ba the opener, sent Liverpool’s odds crashing from 80% to 48%.

Equally shocking was the 3-3 draw away to Crystal Palace from a commanding 3-0 lead late on, but perhaps one of the more overlooked results of this stretch was City’s 3-2 victory away to a strong Everton side who finished on 72 points under Roberto Martinez - still their highest total since winning the league themselves in 1987.

This win improved City’s title probability by more than 30% - almost the same impact as Liverpool’s loss to Chelsea, but the latter is remembered as being far more decisive.

City closed out the title with routine home wins against Aston Villa and West Ham, and a 4th league championship for the club was secured (they now have ten!).

Meanwhile, Liverpool would have to continue the wait for #19, and Steven Gerrard ended his otherwise fantastic career with none. I wonder if he still has sleepless nights about that Chelsea game…

2. Liverpool 1988-1989 (Probability: 93.1%)

The 1988-1989 season is defined by two seismic events, quite different in nature. The Hillsborough disaster is obviously one, and the other is the climactic final game that is unlikely to ever be surpassed.

Of course, the events at Hillsborough had an enormous effect on Liverpool’s end to this season – not just emotionally, but the task of playing 8 games in May to make up for postponements.

It was actually Arsenal, built around young academy products Tony Adams, Paul Merson, David Rocastle and Michael Thomas, who initially looked like they had blown their chance. Eight points clear on the 6th of May, the model rated them as 75% favourites before a surprise 2-1 defeat at home to Derby thanks to two goals from future Liverpool striker Dean Saunders.

With Liverpool winning 2-1 at Wimbledon on the same day, the odds had flip-flopped, and when the Reds thrashed West Ham 5-1 they entered the final game with a superior GD by four goals. To win the title, Arsenal would need to win by two clear goals in the decider at fortress Anfield, rated by the model as just a 7% probability.

And the rest is history. Early in the second half, a glancing header from Alan Smith gave Arsenal the lead, and incredibly, Thomas got the decisive second with almost the last kick of the game.

It’s hard to envisage a more dramatic climax to a league season. While Sergio Aguero’s winner for Manchester City technically came later (93:20, compared with Thomas’ goal at 91:22), the fact that the protagonists were playing head-to-head surely sets this one apart. Imagine if it had been disallowed by VAR.

1. Manchester United 2011-2012 (Probability: 95.4%)

United, Liverpool, Liverpool, United. If there’s one thing we can learn about teams who blow titles, it’s that the most successful clubs are going to be the ones who give themselves lots of opportunities to mess it up.

We’ve covered United’s 1992 collapse, and it feels appropriate to end with 2012 as those two years nicely bookend the dominant Ferguson era.

I remember watching City fans doing the Poznan during their 6-1 win at Old Trafford and thinking “good for them” - some payback for a lifetime of sitting through garbage while their local rivals win everything. These feelings quickly wore off of course when Abu Dhabi’s billions started hoovering up trophies on an annual basis. But it did feel fresh for a short time.

The dramatic nature of Sergio Aguero’s injury time winner in the final game has perhaps allowed many of us to forget just what a commanding position United had allowed to slip away. Eight points clear with six games to go, United had a 95% probability of retaining the Premier League trophy according to the model.

Even a shock 1-0 defeat at Wigan didn’t seem too terminal, but it was United’s 4-4 draw at home to Everton, having led 4-2 with 10 minutes remaining, that suddenly put City within striking distance as the foes squared off the following week.

A Vincent Kompany header gave City a 1-0 win, and in the blink of an eye they were 60% favourites.

City still faced a tricky visit to Newcastle, who ended up 5th under Alan Pardew for their best finish in a decade. A Yaya Toure brace secured the points, and it now seemed like almost a formality that City would beat relegation-threatened QPR in the final game to seal the league.

But football can be unpredictable, and QPR found themselves 2-1 up going into injury time, after Joey Barton was sent off and had a meltdown that would result in a 12 match ban.

United did their part, winning at Sunderland, and you all know what happened next of course, with Dzeko equalizing and then Aguero driving home the winner to provoke wild scenes at the Etihad Stadium.

Remarkably, this means that United’s title-winning probability dropped from 95% to 12%, then back up in the 90s with City trailing, and finally back to zero. What a ride!

It was also a roller coaster season for City forward Carlos Tevez, formerly of United, who was banished to Argentina following a dispute with Roberto Mancini during a Champions League game in Munich. Eventually the relationship was repaired sufficiently for Tevez to return and start the final six games of the season, City winning all six.

And so City won the title on goal difference, the only time that has happened in the Premier League era. That 6-1 win certainly came in handy. Fergie signed off with one more championship the following year before retiring, and for United it’s been pretty much downhill ever since.

Honourable mentions

6. Everton 1985-1986 (78.0%)

7. Wolverhampton Wanderers 1959-1960 (77.9%)

8. West Bromwich Albion 1953-1954 (76.1%)

9. Ipswich Town 1980-1981 (75.7%)

10. Leeds United 1970-1971 (73.1%)

In my previous World Cup article I wrote about the dramatic migration of international footballers to Europe since 1980. This was illustrated by the Argentina squad of 1978 which contained only 1 of 22 players playing their club football outside Argentina; in the 2002 Argentina squad, this had increased to 21 out of 23 players.

A reasonable hypothesis is that this may have reduced home advantage in Europe-based World Cups since players may already be familiar with the culture and climate. And more generally, the modern world of first class travel and private chefs is quite different to the days when teams would take a boat across the Atlantic. So let’s do some exploration: how much is home advantage worth for the host of a World Cup? Does this vary based on who is the host? And has it changed over time?

Obvious starting point: lots of hosts have won the World Cup

If we simply look at the previous winners of the tournament, home advantage seems to be significant. There have been 22 World Cups played and 6 of them have been won by the hosts (Uruguay 1930, Italy 1934, England 1966, West Germany 1974, Argentina 1978, France 1998).

Loosely grouping the World Cups into old era (1978 and earlier) and modern era (1982 onwards) there has only been one host winner in the modern era, France in 1998. This is confounded, however, by the fact that FIFA have increasingly been awarding World Cup hosting duties to lesser footballing nations such as the United States, Japan, South Korea, South Africa and Qatar, none of whom were realistic contenders for the trophy—although South Korea did end up with an unexpected semi-final appearance.

Aside from home advantage for the hosts, there is also the question of home confederation advantage. Do UEFA teams perform better in Europe than they do in South America, for example? Based on World Cup winners, that would appear to be the case as only one CONMEBOL team has ever won a World Cup in Europe (Brazil in 1958), while similarly only one UEFA team has won in South America (Germany in 2014).

Developing an Elo model to forecast international football matches

It is too simplistic of course to just consider the winners of tournaments when trying to place a value on home advantage. 22 events is not a big enough sample to overcome noise, and it doesn’t account for the quality of the hosts and the teams from each continent.

The World Football Elo Ratings website, much like its counterpart for club football, is a well-maintained website that gives a reasonable overall view of current international team ratings.

While a standard Elo system is attractive because of its simplicity and transparency, I wanted to develop it a bit further to account for some of the nuances of international football. I fit my own parameters such as competition weights, goal difference adjustments and the Elo formula denominator, and I used team-specific home advantage ratings instead of a single, global home advantage, so after each result 87% of the rating change would go to the home team’s overall rating and 13% would go to the home team’s home advantage rating.

Similarly, for inter-confederation matches, 74% of the rating change went to the two teams and the remaining 26% was applied to an overall confederation rating. I found that this improved the model because in a closed system such as world football you have the various confederations whose teams largely play against their fellow confederation teams, with limited cross-pollination in World Cups and other friendlies. This means, for example, that if the United States perform badly at a World Cup and their rating decreases, then in theory that will decrease the rating across all CONCACAF teams as the United States return to continental play. But it wasn’t clear whether such adjustments would disseminate quickly enough across all the CONCACAF teams, and with standard Elo I found that weaker confederations were being systematically overestimated.

For a sanity check, here is a comparison of the current top 10 teams on eloratings.net alongside the top 10 teams in my model. Reassuringly, they are the same teams with a slightly different order, and in my model Germany get in the top 10 at the expense of Ecuador.

We can also take a look at how my model differs from eloratings.net across the weaker confederations. The following table compares the rankings of each team that have qualified for this year’s World Cup. My model ranks CONCACAF teams lower and African teams higher, while the difference in Asian teams is mixed.

Which nations have the biggest home advantage?

As mentioned above, I fit a separate home advantage rating for every team based on its results. The bar chart below shows the top 10:

Something that immediately leaps out is that the biggest three home advantages belong to teams who play their home games at a high altitude: Bolivia (La Paz, 13,412 feet), Ecuador (Quito, 9,350 feet) and Mexico (Mexico City, 7,200 feet). Many of the other largest home advantages belong to CAF (African) teams. Conversely, most of the smallest home advantages are found in UEFA teams, with Germany having the smallest of any team and England the 4th-smallest. What could be causing this?

• European nations are closer together meaning less travel distance for away teams.

• Climates in Europe are more homogeneous when compared to larger continents like Asia or Africa.

• Europe has high-quality infrastructure and wealthier football associations, meaning greater comfort for players in travel and accommodation.

I also wondered about the relative quality and integrity of referees in different confederations, which may make them more vulnerable to the more nefarious aspects of home advantage. But that is a tricky area and I wouldn’t want to speculate too much on a subject I am not familiar with. Nonetheless, it is sufficient to say that there are many other unmeasured factors that could be rolled up into the single measure of home advantage.

Has home advantage changed over time?

The following line graph shows the mean home advantage across all teams in the model. It seems like there was an increase in home advantage between 1970 and 2000—possibly reflecting an increase in intercontinental matchups rather than mostly local ones—and has decreased since the turn of the millennium, in line with the general downward trend in home advantage in football.

Does home advantage increase or decrease at World Cups?

We have a well-calibrated model that assigns an individual home advantage to each team. Now we can inspect past World Cups to see whether host teams have generally had a higher or lower home advantage than would be expected from their normal home advantage rating.

Below we can see the mean residual wins (observed wins minus expected wins) for the hosts of each World Cup, where each game has an expected win probability as forecast by the model using the team’s regular home advantage, and draws are counted as 0.5 of a win.

To reiterate: this is a measure of host teams’ performance over and above what would be expected from the team’s home advantage in a normal game.

In the old era, hosts played a total of 57 matches with an average performance of 0.13 wins greater than expected. In the modern era, hosts have played 67 matches with an average overperformance of 0.01, almost bang in line with expectations.

What can we draw from this? Firstly, it seems that home advantage in World Cups is at least as big as home advantage in normal matches. I don’t think that was an inevitable result—it seemed plausible to me that home advantage might have been reduced by the effect of away teams being based in the country for a prolonged period rather than flying in for a single match.

Secondly, from this sample at least, it seems like World Cup home advantage has been getting lower. But this is prone to a lot of noise. At the most recent World Cup in 2022 Qatar produced the worst ever performance by a host nation (average of -0.39 versus expected win probability). But looking at Qatar’s results since that tournament, it’s possible that they were just an outlier as they have continued to look like a really poor team: since the start of 2025 they have lost to Kyrgyzstan, Uzbekistan, Lebanon, Russia, Zimbabwe, Palestine and Tunisia. They still managed to qualify for 2026, thanks to Asia receiving 9 places in the newly-expanded competition format, and the surprising decision to base the AFC playoffs in Qatar rather than the usual neutral venue. I wonder what influenced that decision?

One notable year in the table is 2002 where Japan and South Korea outperformed their forecast results by only 0.02 per game. This is surprising since both of these unheralded footballing nations won their groups and South Korea went all the way to the semis, knocking out Italy and Spain. So how can it only be +0.02 per game?

One of the reasons is that both teams already had a large home advantage from the model: Japan 235 and South Korea 223. As you can see from the earlier line graph, these are above the average, although not on the level of the top 10. According to the model, Japan were actually favourites for all four of their games, drawing with Belgium (this was before Belgium’s golden era), and beating Russia and Tunisia before losing to Turkey in the round of 16. South Korea beat Poland and Portugal in their group as well as drawing with the United States, before they beat Italy, drew with Spain (winning on penalties) then lost to both Germany and Turkey. Collectively they played 11 games with 5 wins, 3 draws and 3 losses, having been favourites for 7 of the 11 games. So it was a good performance from the pair of hosts, but nothing crazy.

The biggest overperformance in a World Cup was from unfancied Chile who finished 3rd in 1962 after beating Switzerland, Italy, the Soviet Union and Yugoslavia. Aside from Qatar, one of the biggest underperformances was from Brazil in 2014; their 7-1 defeat against Germany in the semi-final remains arguably the most incredible game of football I have ever watched.

How do other international competitions compare with the World Cup?

Above we can see a comparison of host performance versus expectation in each of the major international tournaments. In all tournaments, the hosts have performed even better than would be expected from their normal home advantage, with the exception of the European Championship where hosts have performed an average of 0.038 worse. In fact, the overall trend seems to be for a larger home advantage in tournaments where away teams are far from home (Asian Cup, World Cup, Confederations Cup) with a much smaller home advantage in tightly-packed Europe. And this is on top of the fact European teams already have the smallest home advantages in the model.

So could it be that distance from home is a key factor in the amount of disadvantage an away team suffers? We need to be careful because there are other factors at play here—for example, there may be more disparity between teams in the Asian Cup, World Cup, Confederations Cup and CONCACAF Championship when compared to the Euros and Copa America, where away teams are of a higher calibre and have more experience of playing big games away from home.

Do teams have a ‘home continent’ advantage?

One way of shedding more light on the distance-from-home factor is by seeing how teams have performed in World Cups when they are playing in their home continent (but not in their own country). For example, Germany playing in a World Cup in France, or Mexico playing in the United States.

CONMEBOL and UEFA teams have tended to enjoy a continental home advantage, whereas other teams have not. The problem we have with the weaker confederations is a low sample size.

There have been three CONCACAF-hosted World Cups, but since the hosts themselves do not count for this comparison, that only leaves El Salvador in 1970, Canada in 1986 and Mexico in 1994. In addition to the low sample size, you have Canada playing in Mexico which maybe isn’t a great example of benefitting from ‘local’ conditions. Mexico in 1994 did actually have a reasonable tournament, winning a group containing Italy, Ireland and Norway before losing on penalties to the Bulgarian team that would famously knock out Germany in the quarter finals.

Africa has only hosted a single World Cup in 2010, on the southern tip of the continent in their winter which maybe didn’t provide the most hostile environment for visiting teams. Meanwhile the two Asian World Cups have been at opposite ends of the continent. When it was played in Japan & South Korea, for example, Saudi Arabia were almost as far from home as the European teams; similarly, in 2022 the likes of Japan, South Korea and Australia (who moved to the AFC in 2006 to secure an easier qualification path) were further from home than any European or African team.

What does all this mean for 2026?

This has just been an exploratory analysis, so by no means are any conclusions concrete. But here are some of the takeaways:

• In major international tournaments, host teams seem to have at least as big a home advantage as they do in regular games.

• The exception is the European Championship where host teams perform slightly worse than they would in usual home games.

• CONMEBOL and UEFA teams have historically enjoyed a “continental home advantage” when playing against teams from another continent. Other confederations are hard to analyse due to small and non-representative samples.

• It is possible that factors such as distance from home or climate may play a role in the disadvantage faced by away teams.

I would like to drill down on that last point in a future article if I can find the time. But something that raises my eyebrows is that Mexico are one of the hosts of this year’s World Cup and they have the third-biggest home advantage in world football. They are currently ranked 25th by my model and 20th by the eloratings model. When I apply home advantage, they rise all the way to 5th place. If Mexico and England both win their groups then they could be scheduled to meet in the round of 16 in Mexico City, and according to my model Mexico would be slight favourites for this match. Yet, looking at the betting markets, England are about 7/1 to win the World Cup and Mexico are 80/1. Of course, Mexico don’t get to play at home the whole way because everything shifts to the United States from the quarter finals onward, but that still possibly gives Mexico some ‘home continent’ advantage, especially considering the huge Mexican population in U.S. cities that often gives their matches in the States a home feel anyway.

Winning the tournament is a big ask, of course, but it will be interesting to follow the odds of the host nations in individual matches. Currently Mexico are 1.515 on Pinnacle for their opener against South Africa, while Canada are 1.862 against Bosnia & Herzegovina and the U.S. are 1.884 versus Paraguay. My model prices the home wins as 1.18, 1.29 and 2.44.

Now I’m not so naive as to think a simple Elo model is going to be sharper than the betting markets, but it’s interesting to observe the differences and find explanations for them. I don’t think home advantage is being priced in as much as perhaps it should be, which makes me believe there is scope for my Mexico and Canada prices at least being directionally-correct and perhaps offering some value on those two games. The United States are trickier because clearly some amount of improvement is being priced in relative to their poor results over the last couple of years. But Paraguay are significantly more difficult opponents than South Africa or Bosnia & Herzegovina, so it seems wrong that USA should be essentially the same price as Canada for their respective home openers.

Hopefully I’ll be able to add another part to this article where I try to model the effects of travel distance and climate. Either way, I will continue previewing the World Cup alongside the climax of the domestic football season. If anyone has any thoughts on World Cup home advantage, or if you can spot any flaws in my logic or methods, I am an open book and I always welcome feedback so feel free to comment or send me a message.

Brentford 0 - 0 Fulham

A fifth consecutive draw for Brentford who have gone off the boil just as they had a realistic shot at a Champions League place. There’s still an outside chance for them thanks to Chelsea’s recent slide, if Aston Villa can finish 5th and win the Europa League which would give the Premier League an extra CL spot for the 6th-placed team.

Overall it was a reasonable performance from Fulham who shaded the contexG 1.46-1.35 as they dominated territory (12 deep completions to 4) even though Brentford appeared vastly superior on Understat’s xG (2.34-0.79). But Fulham didn’t manage a single shot on target in the match and had Bernd Leno to thank for a fantastic save in the 90th minute to preserve their point. Not a lot more to add about this stalemate.

Leeds 3 - 0 Wolves

Leeds are just a good team. Not “good for a promoted team”; just good. Daniel Farke was pretty close to losing his job after a poor start to the season but he switched formation to a back three at half time away to Man City (match #13) and it changed everything. They have been performing at the level of European challengers and with easy fixtures like this one awaiting them, I would have been shocked if they weren’t able to pull away from the relegation zone as they have now done.

This was a commanding performance from Leeds against a poor Wolves side as they went 2-0 up early and never looked threatened, finally tacking on a late penalty for an emphatic Understat xG of 3.56-0.57. Wolves’ mini-renaissance was really just a mirage thanks to some positive finishing luck and they have turned in two woeful performances against relegation threatened sides over the last two weeks. Next week we will see if they make it a hat-trick as they host Spurs.

Newcastle 1 - 2 Bournemouth

Bruno Guimaraes’ summer transfer fee continues to increase as the Geordies once again looked awful without him in the side. Bournemouth dominated the first half, racking up 2.17 xG and leading 1-0 before Bruno was introduced on the hour mark. Now I’m not saying he changed the game single-handedly but Newcastle had levelled within 7 minutes of his arrival and dominated the remainder of the game with 7 shots to 4, two of those Bournemouth shots being from their winning goal (Fotmob have deemed Evanilson’s header to be a shot rather than a pass; dubious if you ask me).

Newcastle’s stark decline in form does seem to have perfectly coincided with Bruno’s injury and at age 28 this feels like the summer where he will fancy a bigger challenge and Newcastle will know they need to cash in. He’s been linked with Manchester United who are wealthy enough to absorb the lack of resale value, and Bruno seems like a pretty solid investment for a Premier League-ready midfield upgrade. Newcastle fans will be hoping they can reinvest the money a bit more wisely than last summer’s calamitous buys.

Tottenham 2 - 2 Brighton

The bar was very low for Spurs but they produced a much-improved performance by playing in-form Brighton to an even game (contexG 1.56-1.48). Of course, results are the main thing at this stage of the season and the injury-time equaliser was a devastating blow to their survival hopes, but at least there are signs of life on the pitch after a woeful spell under Igor Tudor that looks like it may end up being the deciding factor in the relegation battle.

It sort of sums up Tottenham’s season that after Cristian Romero’s season-ending injury last week it was his replacement Kevin Danso who made the crucial error resulting in Brighton’s equaliser. The weird Xavi Simons saga continued as he finally started a league game, as everyone had been crying out for, and—lo and behold—he gave a man-of-the-match performance with a goal and an assist, also hitting the post, before suffering from severe cramp and effectively leaving Spurs with 10 men for the last 15 minutes of the game.

Brighton will be disappointed they couldn’t take advantage of another Chelsea defeat but they were unfortunate to come up against a much sterner test than the usual dross Spurs have been serving up at home this season. They have the chance to leapfrog Chelsea into 6th place as they host them on Tuesday.

Chelsea 0 - 1 Man Utd

Something of a heist at Stamford Bridge as Manchester United had 4 total shots, only one of which was inside the box, but it went in the net and they survived Chelsea peppering their goal and hitting the woodwork three times to escape with a 1-0 away win.

Chelsea were booed off by their supporters which seems harsh after they had been comfortably the better team against a side who are third in the table. But we are at the point where anyone under 30 doesn’t remember pre-Abramovich Chelsea and just thinks they are a club who should hoover up trophies on an annual basis. These ghosts from great eras of the past can haunt many a football club.

Chelsea are now scoreless in four matches since they thrashed Aston Villa 4-1 at the start of March. They thoroughly dominated this contest (40 touches in the box to United’s 10) and managed 21 shots but couldn’t convert any of them. Liam Delap led the forward line and he now has just one league goal in 24 appearances for the Blues. I still don’t think they are far away from being a very good team if they can find some genuine star quality up front (or just keep Estevao fit) as well as injecting a bit more experience and leadership to guide the young talent. But if Chelsea miss out on a Champions League spot, as now seems likely, it remains to be seen whether they can conjure up yet more accounting shenanigans to avoid having to sell a star player or two this summer.

Aston Villa 4 - 3 Sunderland

A weird game at Villa Park with 7 goals from 25 shots. Really this was a dominant and comfortable home win for Villa except for the bit where Sunderland repeatedly shredded them open for a five-minute stretch at the end of normal time, going from 3-1 to 3-3 and missing a one-on-one with the keeper for a very unlikely victory, before Tammy Abraham popped up with Villa’s late winner.

The story of Villa’s season was their outrageous overperformance of xG in the first half of the season, scoring a large number of long-range goals and continually finding ways to win by a single goal to mount an unlikely title challenge. That was never going to be sustainable, and after the home thrashing by Chelsea it really looked like a CL place was in jeopardy but they have responded well and although this win was sort of typical—scored in injury time to win by a single goal—this result was thoroughly deserved with contexG scoring it 2.22-0.98.

Champions League football is pretty much in the bag now which means Villa can focus on their forthcoming Europa League semi-final against Nottingham Forest which, given Freiburg or Braga await, feels more like a final. It’s clear that England are going to absolutely dominate this competition over the near future which makes the race for European qualification even more enticing given the lure of a trophy and Champions League place to the winners.

Everton 1 - 2 Liverpool

A painful one for Evertonions as the Reds win the first Merseyside derby at the Hill-Dickinson stadium with a 100th-minute Virgil van Dijk goal from a corner. ContexG scored this 1.17-1.67 which, after adjusting for opponent strength and home advantage, is about in line with the season-long performance of both teams: Liverpool in the top 5 but some way off title challengers, and Everton a bottom-half team who have been overperforming the numbers by quite a lot.

Everton went into this game with the chance to move two points behind their neighbours in the table and so the late winner was hugely consequential in solidifying Liverpool’s prospects of Champions League football next season with Chelsea, Brighton and Brentford all failing to win. It was far from vintage stuff from the Reds with only 17 touches in the box, and it’s been far from a vintage year but at least they will keep that truckload of Champions League cash rolling in as they plan for a summer where they will be moving past the Mo Salah era, and possibly the Arne Slot era as well.

One extra note from this game is Liverpool’s backup keeper Mamardashvili went off injured, leaving third-choice Freddie Woodman in goal, but it sounds like Mamardashvili’s injury is superficial and not long-term.

Nottingham Forest 4 - 1 Burnley

A crucial win for Forest and an emphatic scoreline but not a convincing performance against the league’s worst team. The first half was absolutely dreadful and Forest were booed off by their fans having had only three shots and trailing 1-0.

Things turned around in the second half thanks to a Morgan Gibbs-White hat-trick but the final scoreline was very flattering as Forest scored 4 times from just 10 shots. This follows their previous league win at Spurs where they scored 3 goals from 8 shots.

The common narrative is that Vitor Pereira has steadied the ship and pulled Forest away from the relegation zone as well as progressing in Europe. But it feels like that win away to Spurs, who were so poor under the hapless Igor Tudor, is doing a lot of the heavy lifting.

The below chart shows Forest’s recent performances which have not been great according to contexG—ironically, the tall green bar was the draw against Wolves that got Sean Dyche the sack after Forest had 35 shots, the most for any team in a Premier League match this season.

Forest’s Europa League run has also seen them benefit from a penalty shootout win, a bizarre own goal and an early red card for Porto. I’m not convinced anything has fundamentally changed, but they are now 5 points ahead of Spurs and may only need one more win to be safe.

Man City 2 - 1 Arsenal

The big game of the weekend and the season. It has increasingly become clear that City are the best team in the league and they cemented those credentials with a dominant performance over Arsenal.

The shift in the landscape from the start of the season to now can be illustrated by comparing the statistics from the matchup between these teams back in September. Arsenal dominated that game with 39 touches in the penalty area to City’s 7 and 10 deep completions to City’s 3. Conversely, in this weekend’s clash City led penalty box touches 40-14 and deep completions 15-3.

It certainly wasn’t a comfortable win as Arsenal’s rare attacks did end up creating a one-on-one, which Kai Havertz missed, and Arsenal also hit the woodwork twice via an excellent shot out of nothing from Eberechi Eze and a deflected Gabriel header. There was also the absolute sitter missed by Havertz in injury time which will be replayed both literally on TV as well as metaphorically in the minds of Arsenal fans if they go on to lose the title. It was a miss that somehow seemed emblematic of their season: if Arsenal had a forward line even remotely as good as everything else in their squad, they would have absolutely hosed up in the league.

But they don’t have that, they have Gyokeres and Madueke and Eze and Trossard and Martinelli, and Havertz still recovering from injury and Saka currently missing with one. It’s a good set of forwards for most clubs, but when you are challenging to be the best team in the richest league in the world you really need that top drawer talent: players like Erling Haaland, who scored the winner with his 23rd goal of the season.

Only weeks after Arsenal touched 1.11 to win the title (90% probability) we have officially flip-flopped, and City are now trading around 1.72 (58%). City travel to Burnley on Wednesday where a win would move them into first place on goal difference, or goals scored at the least.

Looking ahead, there could be some interesting storylines at the end of the season as the final game sees City face Aston Villa while Arsenal travel to Crystal Palace, so both opponents may be prioritising European finals before or after the game. It would be incredible if goal difference is the decider and one of the opponents fields a severely weakened side, but of course there is a lot of football to be played before then. Every match in isolation looks like an easy result to pencil in but the 11-match accumulator for Arsenal and City to both win all their remaining games is unlikely to come in. It should be an interesting month.

The final game of the weekend sees Crystal Palace host West Ham on Monday night. Palace have eyes on the Conference League and a few injury concerns, so it remains to be seen how seriously they will focus on this game. A win for the Hammers would leave Spurs 4 points adrift of safety.

Because of next weekend’s FA Cup semi-finals, three fixtures have been brought forward to midweek including Burnley v Man City and Brighton v Chelsea. Then the weekend’s slate sees Arsenal try to get back to winning ways against Newcastle, while Spurs travel to Wolves in what is surely a must-win match. Enjoy the football, and look out for my latest World Cup article which I will be putting up in the next few days!

A few weeks ago, I published an article containing the results of my Premier League simulations after Matchweek 31. Although only one round of league fixtures has been played since then, a lot has changed. Let’s take a look at the new numbers.

Again, I have tried to use consensus team ratings based on how the markets are pricing teams in matches. I could use my own ratings but that’s not very interesting because the forecasts will then be driven by whichever teams I have higher or lower than the public.

The title race: flip-flop imminent?

My previous article was published on the 27th of March. Arsenal still held a commanding lead at the top but they had just lost to Manchester City in the Carabao Cup final, and there was speculation about whether that would be a psychological blow. Since then, Arsenal have additionally lost to Southampton in the FA Cup, secured a fortunate late win after an uninspiring performance at Sporting Lisbon, and then surprisingly lost at home to Bournemouth in the Premier League where they created just 0.18 xG from open play.

For City, meanwhile, the trend couldn’t be more different. They followed up their cup final victory over the Gunners by thrashing Liverpool 4-0 in the FA Cup quarter final, and then produced a commanding performance to win 3-0 at Chelsea. My contexG ratings scored this as City’s best performance of the season, on the same weekend as Arsenal produced their worst.

In spite of City now being rated better than Arsenal, the Gunners still win the league in 65% of my simulations compared to the markets which have them around 59%. So why the difference?

Firstly, the vibes factor. There’s no doubt if you polled for public sentiment at the moment, an overwhelming amount of people will feel like the momentum is with City now and that’s going to continue. A casual view would be: you fancy City to beat Arsenal this weekend, and then you look at each of City’s remaining fixtures and you fancy them to win each one. And it’s City, it’s Pep Guardiola: they know how to get the job done, whereas Arsenal have developed a reputation for “the B word”.

I can’t argue with any of that sentiment, but it’s important to anchor ourselves with numbers. The consensus ratings I used have City about 0.4 goals better than Arsenal. If we make Arsenal worse by about 0.2 more goals, which would put them on a similar level to Liverpool, then their title odds would decrease to 55.4%. So it seems like the markets are averaging Arsenal at something like half a goal worse than Man City at the moment. I’ll leave you to decide if that’s fair.

Let’s remind ourselves of the remaining fixtures for the two teams. City’s postponed game at home to Crystal Palace is shown as Matchweek 32 here, but it will likely be rescheduled for something like the 13th of May.

Note that the colours and averages in the above table are taken from my contexG ratings, with a league-wide home advantage parameter. As described earlier, the actual simulations used consensus ratings.

This weekend’s title clash is a huge game, not only because the two rivals go head-to-head but because both teams’ run-ins are so soft that it is clear any lead becomes even more valuable. Using the same consensus ratings that made Arsenal 65% favourites, I plugged in three different results for the Man City v Arsenal game to see how they would change the forecast. A 1-0 Man City win would make Arsenal 47.5%underdogs, a 1-1 draw would make Arsenal 76% favourites, and a 1-0 Arsenal win would make them 93.8% favourites. You can obviously make decent estimates of the relative chance based on slightly different ratings for Arsenal, based on whatever your view is.

It depends how the game goes, and with goal difference potentially in play a score of, say, 3-0 is more significant than 1-0. But my estimate of the title prices after a City win would be something like 1.6 City and 2.66 Arsenal, which would be an incredible flip-flop after Arsenal have traded as short as 1.11.

The race for Europe: a good week for Liverpool

There wasn’t a dramatic amount of change in the European races off the back of last weekend’s results. It was a positive set of results for Liverpool whose Champions League odds grew from 66.8% to 82.3% after they saw off Fulham 2-0 while Chelsea got pumped at home by Man City. Manchester United had a disappointing loss at home to Leeds but thanks to Chelsea’s defeat they are still rated more than a 92% chance to make the Champions League.

Chelsea’s odds have dropped to 28.8% and they are now slight odds-on (51%) to be playing in the Europa League next season. The week’s big winners were Bournemouth, whose shock 2-1 win at the Emirates (and accompanying ratings boost) saw their odds of qualifying for Europe climb from 8.7% to 26%.

Note that these probabilities were calculated prior to Liverpool’s defeat in the Champions League on Tuesday. However, they were only assigned a 2.5% probability of winning the tournament (and thus re-qualifying for the CL, in the event they did not already qualify via the top 5) and so the effect should be minimal.

The two big games in the European race this week see Everton v Liverpool in the Merseyside derby, while Chelsea host Manchester United in an 8pm Saturday night kick-off (still hard to get used to that). Both home teams could really do with a win to boost their European hopes, while both away sides would probably be ok with a draw. If I plug in a 1-0 Everton win, their Champions League odds jump from 3.6% to 10.3% and their Europa League odds from 23.5% to 36.1%. Meanwhile a 1-0 win for Chelsea would increase their CL chances from 28.8% to 40.8%.

Relegation: it just keeps getting worse for Spurs

Just a disastrous weekend all-around for Tottenham Hotspur as they lost at Sunderland with star captain Cristian Romero suffering a season-ending injury, while both West Ham and Leeds won and Nottingham Forest gained a point against Aston Villa.

Using the consensus ratings I now make Spurs a 52% chance to get relegated which is higher than the current market price of around 48%. This is interesting, because last time my simulations thought West Ham were good value. So what’s going on, exactly? There are a couple of possibilities:

1. The market underestimates the value of points on the board. Possible, but I was doing these sorts of simulations 25 years ago, and there are lots of clever people with very deep pockets betting in these markets, so I don’t know if it’s a simple case that punters don’t know how to plug a load of match odds into a simulation.

2. There is an implied extra incentive for teams who are in the bottom three. Imagine you were in a group of people being chased by a bear. You might be running very fast. You might even feel like you are running as fast as you possibly can. But if you suddenly became the backmarker of the group, the closest to the bear, perhaps you might find that little bit extra from somewhere?

In scenario 2, given that Spurs are the biggest club to face a genuine threat of relegation in the Premier League era, it is fair to say that the effect of this incentive is potentially even greater as their struggles are more likely to be due to a lack of motivation and organisation than simply a lack of talent. That’s what Spurs fans will be clinging to anyway, because recent evidence from the eye test would give very little suggestion of an imminent uptick in form.

.The one positive bit of news for Spurs was Arsenal’s defeat to Bournemouth which—aside from local banter concerns—drastically increases the chances that Arsenal will go into their game at West Ham needing a result.

I have been fairly bullish on Leeds’ survival chances given their performances this season and their still-to-play home matches against Burnley and Wolves, and they pretty much sealed their safety with a delicious win away to hated rivals Man Utd. It was an odd game as Leeds produced a fantastic first half performance to lead 2-0 but then ended up hanging on a bit against ten men after Lisandro Martinez was ridiculously sent off for a “hair pull”. Credit to Leeds who have really been a Premier League level club for a few years now but have suffered from narrow promotion misses and the handicap of having Ilian Meslier in goal for multiple years. But it feels like they should be in for a sustained run in the top flight now.

This weekend Spurs face in-form Brighton while Nottingham Forest are at home to hapless Burnley and West Ham travel to Crystal Palace. If things carry on as they have been going then Spurs will be needing snookers pretty soon. But things don’t always carry on going as they have been going, which is why gambling often involves the loss of one’s money. Whatever happens, I’ll be back next week to pick up the pieces and review the new situation.

West Ham 4 - 0 Wolves

A fantastic weekend for West Ham as they moved out of the relegation zone with an emphatic win over Wolves. This felt like Nuno-ball at its best: a fairly even first half with the Hammers going ahead from a corner just before the break, and then in the second half West Ham put on a counterattacking masterclass with 38% possession, two goals on the break and another from a set piece.

There is no doubt West Ham have seen a distinct improvement since Christmas (see below). Last week’s performance at Villa was awful, but aside from that they have generally been playing more like a mid-table side than the shambolic early-season efforts. Perhaps a lesson that new manager bounce sometimes takes a bit of time to bear fruit?

Arsenal 1 - 2 Bournemouth

ContexG has this as the worst performance of the season for Arsenal, surpassing the defeat at Aston Villa in December. The headline xG (Understat 3.16-1.39, Fotmob 2.32-1.19) may give the impression of an unlucky defeat but Arsenal benefitted from a comedy handball penalty and created just 0.18 xG from open play in a game where they never led, playing at home to a Bournemouth side that has conceded the 5th-most goals in the league.

Worryingly for Arsenal this was not out of the blue—for the last couple of months they have been playing more like a team challenging for the CL spots than a title-winner. In the last eight league games only the Spurs match was exceptional, and if we are being honest that game said more about Spurs and Igor Tudor than it did about Arsenal. Throw in the defeats in both domestic cups and some underwhelming performances in soft Champions League ties, and there has to be real concern whether the Gunners are playing well enough to get over the line.

For context, in the eight previous seasons for which I have contexG ratings, the average contexGD of the Premier League champions has been +1.73. Arsenal have reached that level in a handful of games but their season-long average is +1.03 and this year’s PL does seem bereft of any elite teams.

Man City got past their biggest hurdle with a comfy win at Chelsea and that sets up an epic head-to-head next weekend. In spite of everything, Arsenal are still six points ahead and if they win that game it is basically over.

Brentford 2 - 2 Everton

The spoils were shared in a lively game between two unlikely European challengers. A point apiece does no harm to either team’s Europa League hopes but an outside shot at a Champions League place seems even less likely now.

ContexG scored this game 1.82-1.32 in favour of Brentford and they will be frustrated to have conceded an injury-time equaliser. In fairness, both Brentford goals had a touch of fortune about them with Jordan Pickford conceding a penalty by needlessly scything down Kevin Schade as he was heading away from goal, while the second goal deflected in off Igor Thiago.

Everton host Liverpool next week with the chance to go within two points of their local rivals. The Toffees have finished at least 12 points behind Liverpool every year since they last finished higher in 2012-13, and it’s always nice when the big local derbies have more at stake than just bragging rights. Should be a cracker.

Burnley 0 - 2 Brighton

I have been highlighting Brighton’s marked improvement in recent months—coinciding with the re-signing of Pascal Gross—and this run of form continued with a dominant win at Burnley as the Seagulls led for more than half the game and bossed territorially (penalty box touches 15-39; deep completions 3-11).

Aside from the poor performance at home to Crystal Palace, Brighton have been playing at the level of Champions League challengers rather than the ropey form that had Fabian Hurzeler under pressure a few months ago.

Brighton did ride a bit of luck in this game with two Burnley goals being disallowed by inches. It was also an unusual day for Mats Wieffer who had one goal in 51 appearances for the Seagulls prior to today and ended up scoring both goals. But this is a very dangerous Brighton side that still has an outside chance of an unlikely Champions League spot.

Brighton travel to Spurs next week and were slight underdogs prior to today’s significant injury to Cristian Romero. That moved them to favouritism (currently 2.46 on Pinnacle) but that still feels like a generous price given the current quality of their play, allied with Tottenham’s continuing struggles.

Liverpool 2 - 0 Fulham

It’s been a disappointing season for Liverpool but that is only by Liverpool’s standards. They are still able to routinely dispatch a team like Fulham, amassing an impressive 21 deep completions against a side that can be dogged and hard to break down.

The first half was dominant: 30 touches in the box, 10 shots to 5, and cracking finishes from the forwards on opposite flanks and at opposite ends of their careers, Ngumoha and Salah.

Fulham dominated territory in the second half with 31 touches in the box compared to Liverpool’s 17, but they were trailing 2-0 and only managed one shot on target with 8 of their 15 second-half shots blocked. It’s a concern that Liverpool’s midfield is unable to kill off a game more convincingly but it was a big three points in the top 5 race with the Reds still facing trips to Everton, Man Utd and Aston Villa as well as the crunch clash against Chelsea at Anfield.

Crystal Palace 2 - 1 Newcastle

Palace leapfrogged Newcastle in the table thanks to an injury-time winner from the penalty spot after a shirt pull by Sven Botman. There’s a lot you can get away with as a defender but shirt pulling is difficult for officials to wave off—why do they do it?

It was a tight game with only 2.31 total contexG and Newcastle, well-rested and out of all the cups, again left their three expensive forwards Woltemade, Elanga and Wissa on the bench. Amid some stiff competition from other clubs, Newcastle’s summer 2025 window looks especially disastrous and could set the club back years. It remains to be seen whether Eddie Howe will be back next season as the fans who made the long trip to Selhurst Park are surely losing patience with the team now lying in 14th place and unlikely to be playing European football next year.

Nottingham Forest 1 - 1 Aston Villa

Despite starting the day 22 points behind Villa in the table, Forest went off as the betting favourites for this Midlands derby, reflecting a general perception of Villa’s overperformance this season as well as Forest’s improvement under Vitor Pereira.

ContexG scored this game 1.26-1.42 so it was a reasonable away performance from Villa as Youri Tielemans made his first league start since January. Given results elsewhere it can be seen as a decent point for both sides as it moves Forest further away from Spurs at the bottom, and Villa are now seven points clear of Chelsea in the top 5 race.

Sunderland 1 - 0 Tottenham

A new manager but it felt like more of the same for Spurs with some bad luck, a key injury and lots of toothless attacking play.

It was a tight game that always felt like the first goal would be crucial, and Sunderland scored what proved to be the winner thanks to an enormous deflection on Nordi Mukiele’s shot. This was followed by a bad-looking injury for Cristian Romero who went off in tears, no doubt fearing his season is over and World Cup in jeopardy.

In truth, Spurs never really looked like equalising and their contexG score of 0.86 reflects the fact they could only conjure three shots for 0.12 xG after going behind. It doesn’t reflect well on Xavi Simons that yet another manager has watched him in training and decided he wasn’t worth introducing until the 85th minute despite Spurs crying out for some creativity.

With West Ham thrashing Wolves it was about as bad as the weekend could go for Spurs. I mentioned in my recent article that I felt like if you wanted to oppose Spurs it was better to do so in individual matches than in the relegation market. And I think this point is well-demonstrated by this weekend’s results where the West Ham-Sunderland double would have paid more than 4/1 and yet Spurs are still odds-against to go down. I will try and get an updated simulation out this week but it seems like anyone who opposed Spurs in the relegation market has an opportunity to escape with limited loss in a situation that is now drastically worse than it was before these fixtures.

Chelsea 0 - 3 Man City

What a huge weekend at both ends of the table. Next week’s title six-pointer against Arsenal looms large but this was by far the toughest of City’s other fixtures and they won the game with three goals in a 17-minute spell just after half time.

This was more like peak Man City as they comprehensively outplayed Chelsea on their own pitch: 64% possession, 46 penalty box touches to 14, and 15 deep completions to 2.

I wrote in the aforementioned simulation article that even if City won out, Arsenal would still win the title a third of the time because the Gunners still needed to slip up in another fixture, none of which looked especially difficult. But with Arsenal losing to Bournemouth the title is now in City’s hands as well as Arsenal’s.

Season-long ContexG Ratings

This week I included each team’s contexGD in the second half of the season to illustrate how they are playing in recent months. It is clear that Man City are now the best team in the league while Brighton, as mentioned earlier, are now performing like Champions League contenders.

Aston Villa and West Ham have also shown significant improvements in performance compared to the first half of the season. Trending in the opposite direction are Newcastle and Spurs. (Note that the ratings in this table are not opponent-adjusted.)

There is still one match to be played on Monday as Manchester United host Leeds in a fixture that always has plenty of needle between the fans. United (Manchester, that is) are pretty much home and hosed in the top 5 race while Leeds will see this as something of a free hit as they still have home games against Wolves and Burnley up their sleeve.

Next weekend is of course the big Man City v Arsenal clash as well as Everton v Liverpool and Chelsea v Man Utd. I will be back next Sunday with my review, but right now it’s time for one of my favourite days of the year: Sunday at the Masters! Enjoy.