I’ve been meaning to write about squad construction and the peak age concept in Minor Leagues for a long time now, but haven’t had the right mixture of time and tools to do so. This has changed a little bit in the past year or so, as I have put a lot of time into ironing out some creases concerning data storage and ETL, and developing a possession value model.

Possession value models are attempts at evaluating different events based on how they impact the probability of scoring or conceding a goal. There are several frameworks out there (like StatsBomb’s OBV, Karun Singh’s xT, VAEP or ASA’s Goals Added), with some differences in methodology. The fundamental idea remains the same however: based on historical data a successful pass from location A to location B increases the probability of scoring by X, we divide X among the contributing players (passer, receiver etc) in some fashion.

There is obviously a danger in focusing too much on this kind of model: an as yet (mostly) unknown proportion of value on the pitch is created by off the ball actions, so results should be taken with a pinch of salt. However, used cautiously, this can be a very valuable addition to the analyst’s toolbox, in different contexts. The model I chose to emulate was Goals Added (also known as g+), as the folks over at American Soccer Analysis posted a detailed discussion about the measure as a whole, but also about the methodology more specifically. It spoke to me for many different reasons, and one of the major ones was that it isn’t just a framework for evaluating possession actions, but also takes defensive actions into account.

One of the foremost contexts in which g+ can be used, is comparing players between positional categories at the top level. It isn’t quite WAR, but it does something similar in anchoring the language around something concrete that is fundamentally important to the sport, namely goals (or expected such). This means, at least hypothetically, that you can use the same currency when evaluating the impact of a centre forward as a centre back: the value of their actions on the pitch measured in goals. In fact, ASA even have a goalkeeping module, which translates goalkeeping events into g+. I haven’t gotten as far yet, although most of the building blocks are in place.

Squad construction is largely a question of resource optimization, in which the party doing the constructing needs to balance a large array of factors to create a best possible fit. These factors include: budget, playing minutes, short term squad quality, long term squad quality, squad cohesion among others. A successful squad construction is one that fits into the organizations short term objectives (e.g. league position) while remaining on track to hit long term objectives without overextending the organization financially.

A factor that complicates this thinking in Finland is that long term objectives are usually vague, as team finances tend to be on wobbly ground which directly affects the length of personnel contracts (both players and backroom staff). It is also something that is emphasized less at board level – where there may be some overarching ‘goals’ for the long term that mostly lack any measurable link to the other relevant factors (budget, short term objectives etc). The effect of this is that long term objectives become pie in the sky entities separated from short term objectives that are dismissed post hoc as irrelevant – among these are usually ‘making money in the transfer market’ or ‘getting into Europe’.

This means that, as teams in Finland live hand to mouth, the bulk of squads are usually built with the upcoming season in mind, rather than anything beyond that. It’s partly a necessity due to the labile nature of Finnish football, but is also partly due to a lack of focus on the longer term and poor talent identification.

Squad construction analysis is essentially built around the footballer aging curve, which tends to look something like this:

Although there is some positional variance, players tend to have quite a steep development curve up until their early-mid twenties after which they plateau for a while, and then head into a slow decline as they age. This is important to know, and take into account, as a team is built to fit its short- and long term objectives, because age can be used to approximate future player performance. If we decide that we want to be competitive within three years, the most prudent strategy might be to bulk up on pre-peak players, allow them to gel together as they develop, and hit their peak in a couple of season’s time. If we’re expecting to be competitive immediately, signing peak-age players could be considered smart thinking as they are likely to be more productive than their counterparts on either side of the age distribution. Player salaries also usually follow a similar trajectory, with younger players being less expensive than peak-aged players, which means that if you want to build a team on a budget, having an eye on the squad’s age profile is a good idea.

This concept works well at the top level and in the aggregate. Compare two completely random players, one being 20 and the other being 26, and it is highly likely that the latter will be better. What complicates this thinking is when you drill down to a more local level and apply existing market restrictions. For example, for a player in Finland, there is a ceiling for how well they can play before their perceived value on the global market will surpass their internal valuation, and if the player stays above that ceiling for a long enough sample of playing time, they will essentially remove themselves from the sample – at least for the time being. This is affected by two additional things: contract duration (pushing their internal valuation down as time left on the deal shortens) and player age (the older a player is, the less enticing the foreign opportunities will be). This means that while the individual player development curve will regularly have the expected look to it (rising until it hits mid-20’s, then stagnating and falling as the player gets into their 30’s), as a whole, market forces will create a different picture.

Due to the position of the Finnish market, access to known high quality peak age players is essentially minimal. If a foreign player is good enough, there will be some other reason for why they are accessible (injury, personality, something else). If the peak age player is domestic, there will be some reason for why they aren’t playing abroad as that is what the vast majority of professionals in Finland dream of – even going to the lengths of moving to foreign lower tiers to realize this dream.

A contributing factor to this effect is the high standards of living in Finland, and the relatively low wages on offer for professional footballers. For a young player, making 800€ a month can sound like a decent proposition because the alternatives are school or a better-but-still-low-paying, less interesting job. The older a player gets, the more difficult it becomes to justify hanging on to those final hopes of realizing the dreams of a professional career. Other countries also suffer from the same thing, with Norway having such high entry level wages for menial jobs that many young players, at an even earlier age, opt for the safety of a steady paycheck over the career rollercoaster of professional sports. There is naturally a trickle down effect, with players who get cut at higher levels going down levels to work their way back up, but – again – it is difficult for any peak age player to justify moving to a different country to play for 1500€ a month, unless that money is significantly more than they can make where they reside at the time.

Basically, for any individual league, an age curve will ultimately be strongly affected by player development, but also by availability. This will naturally also have some implications for people trying to construct squads in this environment.

If you look at the picture in the embedded tweet, you’ll notice that the y-axis lacks a label. The measure being displayed is probably some type of player quality measure – something in the same ball park as goals added, for example – but could also be something like a playing time distribution. At the top level, minutes played is a decent proxy for player quality as generally, good players tend to get picked over less good players. If you look at the playing time distribution for what I have dubbed the ‘Big leagues’ (the top tier in England, Spain, France, Italy, Germany and the Netherlands), you’ll notice that it follows a similar pattern. Note that I haven’t plotted minutes played, because the maximum amount of playing time available for a player is dependent on the amount of matches they play, rather, I have taken the proportion of the maximum available minutes that each player has played, and plotted the mean for each ageseason.

Playing time distribution in the big leagues, data from transfermarkt.com

A steep curve followed by a plateau, and a slow decline. Let’s have a look at the same graph for Finland (top two tiers).

Playing time distribution in Finland’s top two tiers, data from InStat

This doesn’t look at all like the previous picture. This looks more like a steep increase, and then a slightly less steep increase which doesn’t seem to tail off at all. Let’s compare the two by overlaying them. The Finnish data only contains players with over 100 minutes of action so we’ll also add that condition to the Big leagues data here.

Playing time distribution for Big leagues versus Finland

The plots seem to follow the same trajectory roughly until age 24 which is when the plateau starts in the Big leagues. The Finnish plot also hits something of a plateau but there is a slight growing trend that continues even as the other plot starts its decline. The difference between the plots keeps growing as we get into the 30’s. There are several potential reasons for this effect. For one, the higher the level, the more physically taxing the game is, which will likely start to push players out as they age. This trickle down effect essentially leads to (mostly domestic) players coming to Finland late in their careers and being capable of carrying a higher workload than for their previous clubs. There are also fewer minutes in total in a Finnish season, so it could be considered more reasonable for any player to receive a larger share compared to the top tiers where there are more games. There is also a survival effect on display. Since wages are low in Finland, players likely retire earlier rather than stick around to play reduced roles, leading to older players being generally of higher quality as ‘survival’ in itself is a signal of some ability. Also, significantly, there is a sample size consideration in the older age categories.

I think it’s interesting to note that Finnish teams look like they are more reliant on peak-age/older players than teams in the Big leagues, as this doesn’t quite track with the idea that there would be a peak age gap. Let’s have a look at player quality as measured by goals added, then.

Goals added performance per ageseason in Finland

It looks similar to the playing time distribution but the slope seems far less steep. There might be something of an optical illusion in play as well, as the bumps in quality after age 31 make it seem like there would be almost linear growth throughout, but if we only look at the preceding ages, the profile looks more like expected. Let’s compare this graph to the playing time distribution. Essentially this can be done by comparing the median for each age category to the full sample median, essentially giving us a plot that tells us how ‘quickly’ a player can be expected to reach median level in both playing time and playing quality.

Player quality of performance versus quantity of playing time in Finnish top two tiers

What we can see from this graph is that the distributions are fairly aligned, but with an increasing gap between ages 24 and 33 (apart from ages 29 and 30). This essentially tells us that in this age range the median player is getting more playing time than their median quality would indicate that they deserve.

So what we have is a situation where on average players in peak age and beyond seem to be utilized more than the quality they produce would dictate, an effect that dissipates and almost reverses as we reach the latter years of a player’s career. The average is instructive of course, but it’s worth having a look at the full distribution of playerseasons, because the edge cases can also show us something.

Goals added playerseason distribution

So in general, I think we would expect the tops of the distribution to follow the average more or less accurately, but that isn’t the case here. Essentially the best playerseasons come from players in the 21-24 age range as well as 32-34, whereas the peak ages in between are quite a lot lower. It’s also worth noting that from age 22 onwards the minimum stays relatively similar but starts to grow as the players age. Essentially, the older ages get their highish average quality from a tighter distribution and the younger ages have higher variance but contain some of the best ageseasons.

In summary, I think it’s fair to say that there is a weird skew in the Finnish age curve, whether you look at minutes played or some other measure.  I think this could be considered a big problem for Finnish football – a league’s peak age players will essentially be what it is built around. One could argue that player wages are a factor. If players could see themselves having a good career making a decent living playing domestically, maybe they wouldn’t have to glance abroad at the end of every season. It also would encourage younger players to keep playing even if they realise that they might never reach the levels they’ve dreamed about. At the moment, it feels like each season has a player retiring pre-peak despite having posted decent numbers previously.

A league that struggles to retain its peak age players is a league in trouble. Peak age player availability can be displayed with one further graph, by measuring the proportion of the population in each age group.

Amount of players per age category in the Big leagues versus top two tiers of Finland

Lest you think that this is merely due to including a couple of seasons of Klubi 04 and one season of SJK Akatemia, let’s have a look at the same graph without Ykkönen.

Amount of players per age category in the Big leagues versus Veikkausliiga

What the graphs show is that there is a dearth of players in the peak age range, something which is corrected at age 31.

This has several implications, chief of which being that recruitment and retention of peak age (or pre-peak age) players isn’t working in Finland. Retaining good Finnish players is practically going to be impossible as long as wages are at the level they are as the temptation of moving abroad is just too large – as it is right now, there are players who accept similar wages from abroad just to get a chance to try it out. Finnish players will make up the larger part of the sample, and access to peak age players is going to be difficult by default, but I wonder if the bigger issue isn’t that we aren’t able to add good (for the level) peak age players from abroad. Just as players from Finland will go anywhere to get a move to a foreign team, I wonder if there aren’t markets where Finland could be that destination. By having a scattergun approach, relying on intermediaries and established markets, Finnish teams by default get the absolute bottom of the barrel. I wonder if the better approach wouldn’t be to establish a presence in some of the markets where Finland would be a genuinely good next step, some of which could even be considered growth markets for footballing talent – think Balticum, Iceland, Faroe Islands, some of the Asian/African leagues, Canada, USL – and then put an effort into talent identification. Even if the median Faroese player isn’t good enough for Ykkönen, the top 1-2% would certainly be good additions for basically any Finnish team (just look at Petur Knudsen, for example, who was shopped around Finland last January but ended up moving to Denmark). The big issue, obviously, is finding out who the 1-2% are, but that’s mostly just a matter of putting in the hours.

For this, one could follow the Canadian Premier League model, of having centralized talent identification, a system that has worked well for the teams in the league. Basically, rather than the teams having to spend resources they don’t have on identifying players to recruit, the league does some of the job for them, amassing a scouting pool from which they can select the players they like. Essentially, this kind of system would mostly make teams less reliant on intermediaries when exploring options, and would help teams with fewer resources become more proactive in their player recruitment.

For a team in Finland, there are some further interesting implications. First of all, if you want to be competitive, it doesn’t necessarily make sense to load up on peak age players. Partly because there simply aren’t that many of them, partly because they tend to be more expensive than they are good, and partly because they probably won’t want to sign on for more than until the next transfer window if they are any good. Aging returnees are a highly coveted segment of their own, but they are already out of reach for most of the market anyway. Would Tim Sparv, Jukka Raitala or Joona Toivio have signed for anyone other than HJK? And if the answer is yes, how many teams could have afforded them? The segment of the market where there is value to be found tends to be pre-peak. I’ve been outspoken in my support for the Wiss-era Ilves squad building strategy, mostly because I think it’s been the most consistent and cohesive. I also think that among all the Finnish teams in the past decade or so, they are the ones who have managed to produce the most with the least, and will remain very underrated. The average age of the squads they produced is one thing, but the consistency of performances is the thing that is really impressive. So for a team that can’t afford aging returnees, the optimal strategy should be to build around players in their early twenties with some signs of competence (either domestically or from abroad) – naturally if you can sign a good player, you should, irrespective of their age, but consider the long term implications and whether it disrupts other parts of the puzzle. If at all possible, try signing your good young players to longer contracts, and try to do it at first sign of a breakout rather than the season after.

For a team that can sign aging returnees, it is a valid strategy to do so. It is, however, worth bearing in mind that while the age curve for Finland might look skewed in the aggregate, individual players still go through the same symptoms of age related decline, so building a team around aging players might be destructive in the long term, especially if getting the player now means signing him on for additional years on similar wages with lesser expected output. Do not align wage with age, if a player is good and young, try to sign them up on a longer contract on proper wages. Try to avoid signing late-peak players from abroad.

And finally, some heuristics when thinking about squad building in a market like Finland:

“When considering a player, try to critically imagine a reason why he would join your club. If the reason doesn’t satisfy you, you shouldn’t sign the player”

“When considering a player, only sign them if you can think of a single realistic scenario in which you can sell them on in one year’s time, even if you wouldn’t want to sell them”

“Signs of some competence at a lower level is more valuable than signs of some incompetence at a higher level”

If you like this post, please consider following me on Twitter. Thanks for reading!


2 thoughts on “Peak-age crisis

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s