You may have noticed a distinct lack of PMR last week! First of all, apologies. As I'm writing this, I'm looking back on the notes that I had prepared for last week's newsletter, which consisted of one job posting and a conference-related update. This was the reason for the lack of a newsletter — I had almost nothing to talk about. One of my friends at Willamette once asked me "do that many things really happen in soccer analytics on a weekly basis?" It's a good question, and evidently, enough happens that I've been able to write weekly newsletters for the last three months. But often, I find myself scrambling to come up with topics or digging into something that isn't always the best fit for the newsletter, simply to pad the word count and make it worthwhile. I don't like doing that, and I'm sure you all have enough in your inbox to read, so I'm going to go ahead and try writing PMR in a bi-weekly format from now on. I will still aim to get all the same updates into the newsletter, but it will be a little less often and, hopefully, a little higher quality. In the intervening time between editions, certain things like job postings can easily get filled, go away, or be buried in the LinkedIn feed. For that reason, any job postings I see will get added to a mega-thread over on bluesky, which you can find here.
The Rundown
(in no particular order)
- 1.
Hudl have announced their talks for the upcoming Hudl Performance Insights/StatsBomb conference in London, with a list of featured speakers here.
- 2.
NFL announced their Big Data Bowl for 2027, with the theme focusing on the relationship between combine data and in-game tracking data. While it is the wrong kind of football, that theme seems especially applicable to translations between academy/youth performance metrics and professional metrics. Keep an eye on the competition, especially since past winners have gone on to excel in all sorts of sports, not just American football. Plus, Nick Wan had an especially insightful part of his thread on bluesky, which I'll just link below for you to read:
- 3.
At ASA, Kieran continued his run by looking at how weather actually impacts soccer games. Lots of discourse has been had about MLS' upcoming calendar switch, with many arguing that the increased cold during matches could ruin the fun, but Kieran finds that the summer, not the winter, is actually one of the biggest detriments to good games.
- 4.
At FAM, Arman dug into the evolution of MLS injury reports, and found a concerning rise in vague-posting. Teams are now using the terms "lower-body" and "upper-body" to describe all sorts of injuries, with use of these terms jumping 5x in just two years. Plus, more teams are using "questionable" to mean "doubtful", especially when a player is a new addition to the injury list.
- 5.
Though not strictly data-related, my friend Jacob Mitchell made his journalistic debut in The IX this week with a piece on injuries in the NWSL. I'll give him points for using one of my radars.
- 6.
The call for proposals for the 2027 ASI Summit closes this week at 11:59 PM CT on Friday, October 23rd. I have been involved with this research competition for two years, both as a competitor and as an organizer, and can't recommend it enough as a golden opportunity for students to get their work in front of incredible people in the industry. There's an open track for non-students, as well. If you were thinking about it, now's your chance.
- 7.
CIES published a ranking of clubs based on how much players appreciated in value while at the team, and while most of the top clubs are the usual suspect, interestingly, Dortmund rank fairly low. Plus, Chicago Fire eeked out AC Milan — thanks, Jhon.
- 8.
A couple weeks ago, Mark Thompson put together a newsletter on the PySport Analytics Cup and accompanying data release of SkillCorner body pose data, which is a major addition to the library of publicly available data that exists today. His analysis of what this data release means for soccer, and inclusion of links to work that has already been done with the data, is well worth the read.
The closer look: Player Projections
SkillCorner posted some key findings of an article by Liam Bailey from earlier this month on LinkedIn this morning, and I couldn't help but be intrigued by the cost-value proposition the post raised. The analysis itself was centered on repeatability — that is, how often a player performs well in the season following a move, depending on how many "good" seasons they had prior to the move — and affordability — how much does that player's value appreciate after each consecutive "good" season. The central question the post asks is this: "Would you rather pay big money for a 74% chance of success, or back several players at 44% each?" More interestingly, however, is how long one should wait before they take that risk. After all, when we have cost appreciation and odds of repetition, this mostly becomes an economic opportunity cost question.
For example, let's use the numbers the charts give us. (I guessed on the 24-27 and 28+ lines, since they didn't list clear numbers for those).
Let's take an arbitrary player who begins with market value $2 million. If this player is under the age of 23, the math is fairly straightforward. Let's assume the player just had her first top-20th percentile season, ranking her in that second row. From the position of a club who is interested in signing the player, assuming the transfer is possible, we are now looking at a $2.1 million player thanks to the 5% appreciation. The odds this player would repeat her "good season" is 44%, meaning that we can expect 0.44 "next good seasons". Of course we don't know the numbers for after that, but we can assume that given her age, the same odds would continue, so an initial investment of $2.1 million secures the following season odds:
Odds of a good season 1 year after transfer: 0.44
2 years after transfer: 0.44*0.55 = 0.24 + Y_1
3 years: 0.24*0.61 = 0.15 + Y_2
4 years: 0.15*0.70 = 0.10 + Y_3
5 years: 0.10*0.74 = 0.08 + Y_4
In these calculations, the numbers we know are only the probabilities that a player will have a good season following a stretch of good seasons, so we add the Ys to account for the probability that a player has a good season despite a broken streak of non-consecutive good seasons. If we knew Y, it would be a lot easier, but we don't, so let's continue, and assume it to be non-negligible. Here, after one good season, we are spending $2.1 million for an 8% chance that the player has five consecutive good seasons following that move, and an overall number of good seasons expected to be 1.01 + Y. Now, let's consider what happens if we wait another season before making the move. In this case, the player's price has now increased again to $2.58 million. Let's do the math again:
Odds of a good season 1 year after transfer: 0.55
2 years after transfer: 0.55*0.61 = 0.34 + Y_1
3 years: 0.34*0.70 = 0.23 + Y_2
4 years: 0.23*0.74 = 0.17 + Y_3
5 years: 0.17*0.74 = 0.13 + Y_4
Chance of 5 consecutive good seasons: 13%
Total expected good seasons: 1.42 + Y
If we repeat that math, we get the following table, with an added column that tells us the price we pay per expected good season:
Of course, this math isn't perfect because we don't know Y, but we can confidently say that Y isn't a constant — it depends on how many good seasons the player has had in the past, even if they weren't consecutive, injuries, how far the player moves, how old the player is, etc. With more time and more information, we could begin breaking off pieces of Y and defining them in terms of each of the above factors. That being said, there will always be some aspect Y that is inherently undefinable — after all, that's the nature of the sport. Because of all that, I know that this process isn't perfect, but rather, this exercise is helpful just as an example of what a process could look like with more information.
Anywho, with the caveats out of the way, we can begin looking at which of the above options makes most sense. Perhaps unsurprisingly, the answer lies somewhere between early action and late action. The best bang-for-your-buck choice is signing a player once they've exhibited 2-3 consistently good years, with the best value proposition at 3 years. That tradeoff balances both price and quality, with the price increase after 4 years reaching almost $1 million. The article raises another option: take multiple swings at a lower cost. While this is a valid approach, teams certainly shouldn't be swinging on unproven players, since the analysis shows that the odds these players work out is far lower, and more expensive, than once they have proved themselves.
For 24-27 and 28+ year olds, the math differs slightly in terms of cost tradeoffs, but the performance math remains the same. Here are the respective tables:
As you can see, for 24-27 year olds, the tradeoff is completely different. In this case, waiting until the player has more consecutive seasons is preferable — though this process is slightly different. While with teenagers, waiting is often easy to do, 24-27 year olds are entering their prime, and must already have this number of consecutive good seasons. That means that costs may be higher, and acting in the moment is preferable. The finding thus suggests that sticking with players with more good seasons under their belt yields better future results and gives you more value for the cost, even if the cost is higher. Essentially: pay for quality, it's worth it.
With older players, the strategy remains the same: pay for experience and quality. In this sense, taking a swing on a less-proven 28-year-old is often less preferable than paying somewhat more for an experienced 30-year-old, despite the age difference. Of course, this is all a theoretical analysis, and as shown above, this math hardly takes into account the myriad other factors that impact how a player will perform in the future.