Valuing trades, draft picks and players – a review of the theory

Last year HPN tried to answer an age-old question in the sporting world: what is a player really worth? This post will review the theory behind what we’ve been doing. It’s rather long, but should give a pretty thorough understanding of the key points.

We have been using a method evaluating trade scenarios by comparing draft picks and players using a common currency to enable direct comparison between the two. The idea is to support discussion of trades and get some idea of what might be over- or under-valuation, and see where clubs might be taking risks or being cautious or unreasonable in how they’re making value judgements.

To do this, HPN created a series of formulas for evaluating the value of current players, young players and draft picks. Even through the first month of deployment HPN tinkered with the formulas, and it remains a work in progress.

The common currency used is remaining career games output and our method of quantification had two main components, one for picks and one for players. In both cases the goal was to determine expected future output from them in terms of games expected.

This off-season we are running a similar analysis again, with new modifications. This post will look at some of the theory behind what we’re trying to do.

  1. Why a common currency and why “future games played” as the basis

The fundamental goal of AFL list management is to use limited resources to assemble the best possible playing list.

List management resources are limited by the salary cap, so clubs are unable to use raw financial power to assemble their playing list in the manner of European soccer clubs or American baseball clubs.

However, resources available to assemble teams are also limited fundamentally by the draft system. The existence of an almost all-encompassing draft limits clubs’ abilities to acquire proven talent by any other method. There are no true zone selections, and free agency remains a limited niche option for improving the list, accessed by already established players. The only real “free hit” is on unproven international talent via the Category B rookie list.

All this turns mainstream player acquisition opportunities, ie draft picks, into a limited commodity, one which must be spent wisely. Those limited opportunities need to be used to get not just a few stars, but to assemble a large body of players, and to ameliorate the risk of selecting complete failures. AFL lists contain between 38 and 40 senior players, and from that, clubs must field 22 players for 22 games. Clubs therefore need their playing list to provide 484 regular season games every year at the highest standard possible.

Anyone capable of playing regular games has inherent value. Playing a lot of games is itself a significant sign of player quality.

Including the first week of finals, 197 AFL players played 20 games or more in 2015. And from there, the drop is steep – 421 players played 11 games, 665 players played at least one game this year. Even the most ordinary, frustrating plonker getting games every week in a best senior 22 is going better than at least two thirds of the AFL’s current player base due to both being selected and not being injured. To put that another way, Jesse White is quite probably in the top third of the AFL system. This is also the logic on which Jake Melksham is worth a decent pick.

  1. Valuing draft picks

We start from this proposition: Given that recruitment is about filling lots of player slots every week, the high value of pick 1 is not just that they’ll give you really good games, but that they’ll give you lots of games, usually almost immediately. Pick 1 is the lowest risk pick; on average, players taken there play the most games. High output by pick 1 players is certainly a demonstration of that those players are typically very good, but as argued above, one of the functions of high player quality is that it provides lots of games for a long time.

For our purposes, this means each pick can be converted into valuation of expected career output based on career games played:

pickchart

There’s some manual adjustment from the polynomial trend line reflecting that some players are still playing (eg Brett Deledio), and also that we wanted to increase the values of picks 1-3 more in line with other draft pick value systems. The full set of figures are below:

afltradechart

From pick 1 we expect about 235 games of football (and hence 2350 points). From pick 10, 123 games (1230 points) and so on.

Note that compared to other draft pick value systems, there’s not a lot of drop-off in pick value in the second round, with pick 19 worth 980 points and pick 36 worth 820 points. This is because we don’t see a real lot of difference between the careers expected of those two picks. By pick 19 the “sure things” are already long gone and a “long tail” of chancier prospects has begun. A lot of good players get drafted at the back end of the second round, and the “games played” figures suggest there’s not a lot of premium on, say, pick 20 over pick 40. We think this has significant implications for how clubs should be using picks in the 20-40 range.

  1. Valuing players: quantity is its own quality

The HPN approach to player valuation treats quantity of games as the core value, in order to compare to draft picks and analyse trade value. The logic is basically that if you’ve got an entirely average player who might give you 100 more games (based on past performance) they should be worth pick 17, which averages a yield of 100 games per player.

In its crudest form, the formula for valuing a player’s future output is [career resources remaining]*[games per season]*10.

If we take completely average 25 year old, with no Brownlow votes and no All-Australian selections, a player who has played 15 games a year due to form and injury issues, we value them at 990 points. That’s roughly pick 18.

To get that figure, we assume a retirement age of 31 and we expect them to have 55% of a typical 19-year-old to 31-year-old career remaining to them, in terms of output. That’s based on our analysis of age cohorts and output (originally discussed here):

resources

That makes the specific calculation [31-19]*55%*15*10 = 990

We therefore expect them to play another 99 games across 6 seasons (increasing their output from 15 games per year to 16.3 in line with the tendency of players to produce more in their late 20s). They could exceed that and play til 35, they could blow out a knee and retire early, but this is the average scenario based on our data.

A player who didn’t miss games and averaged, say, 21.7 per season, would instead be worth 1425 points, with their suggested durability giving them a value of 142 games.

Such a total is obviously not possible in 6 seasons before the age of 31. This is a good point to remind ourselves that the “career resources remaining” calculation contains both quality and volume measures – goals, Brownlow votes and possessions also increase in players’ late 20s. The projected career resources calculation means players are expected to produce more output in their late 20s, in some combination of more games and better games in terms of goals, Brownlow votes or disposals. In this case, a player already playing 21.7 games per season should be expected to keep playing and play most of the 132 games, but also produce games a bit more valuable than the average player.

  1. Valuing players: quality is also quality, but only a bit

The obvious objection to this valuation method, and it’s a correct one, is that all games and players are not equal. The “average games played” of draft picks should be uncontroversial because they don’t pertain to specific players. But implicitly we know some players are better and more useful than others.

To account for player quality or eliteness, we treat quality of player as a modifier of the value of each expected game played.

We think that while quality is a factor in adjusting a pick or player’s value, the volume of their output is the more important consideration. A player’s capacity to provide a large number of future AFL-standard games is probably more important than their ability to be a Brownlow winner as opposed to a solid contributor. After all, you generally wouldn’t trade any decent 22-year-old with a full career ahead of them, for a 31 year old Gary Ablett – GAJ might give you better games but he isn’t going to give you many of them.

A single number 1 draft pick can never be a circuit-breaker for a poor club by themselves, can never drag a team to a premiership alone – at most they can provide 22 of the 484 games needed by a team. With limited opportunities for recruitment, clubs can’t over-concentrate their resources in a couple of big stars as is possible in a sport like basketball or soccer. Quality and quantity of players are both significant considerations in list management.

To support this proposition, we argued last year that the best player in the league was probably not worth two average best-22 players in terms of what they provide. We pointed out that Champion Data rated the average game by Gary Ablett, at his peak, as worth about 130 “supercoach” or “player influence rating” points. This is 1.7 times the 75 point average game by the average player, according to the official stats providers of the AFL. Ablett only gained about 8% of his team’s possessions, and just could not due to the nature of the game, have the massive impact on team results that LeBron James can (with LeBron getting a third of his team’s total win shares).

As an extreme case: if you trade three decent contributors and several draft picks for Nathan Fyfe, you’ve taken away potentially hundreds of games of future AFL-standard playing capacity. You’ll have Fyfe, but you’ll still need to find ways to fill the rest of your list.

Clubs demonstrate this in their actions, too – St Kilda have sought in recent years to multiple high picks to fill out their list with a critical mass of talent. Hawthorn have showed a preparedness to trade down the draft in order to get two shots at lower picks rather than one shot at a higher pick. And how often do you see three players swapped for one these days?

In short, clubs need to be bringing in a decent quantity of available AFL-standard talent, not just a small amount of quality players. We therefore treat quality as a modifier that makes players’ games output more valuable, but not to the extent that a game by Gary Ablett is worth more than a game by two Ryan Bastinacs.

  1. Valuing players with loading for player quality

The way we load for player quality is fairly straightforward, and done in two ways. As discussed above, we view “future career games” as the fundamental output of comparison between players and picks. So we recognise that quality adjusts the value of a player’s game value by multiplying expected games by the quality loading.

If Gary Ablett is worth, at a ceiling suggested by Champion Data, about 170% of the average player in each game he plays, we can get close to that mark using Brownlow Votes and All-Australian selections over the most recent three seasons:

Ablett got 74 votes in 2012-14, which we would treat as a loading of +74% by applying a 1% loading for each Brownlow vote. He had three All-Australian selections, which we treat as a loading of +60% in that period (20% per AA, 10% for being on the squad). The higher loading of the two is the one we use to adjust a player’s value for their player quality. These figures are arbitrary, but the important thing is they max out somewhere around that Gary Ablett Champion Data high water mark.

The second way we recognise an increased value of future games output by elite players is via a later assumed retirement age. All-Australian selected players tend to retire around the age of 32.4 years, whereas the standard AFL retirement age seems to be more around 31. That extra 18 months of assumed career increases the amount of games an elite player can be expected to provide.

For an example of the impact of eliteness in our analysis, let’s compare that completely average 25 year old who played 21.7 games per year, worth 1425 points, to Patrick Dangerfield.

In an earlier post we gave Dangerfield’s value as 2859 points (which would translate to 286 AFL-standard games, and close to double our generic 25 year old). This is derived by:

Career resources remaining of 60% (calculated from player age outputs which factors in a later retirement age)

Assumed retirement: 32.4 years old (giving a career length of 32.4-19)

Expected games per season: 21.7 (the average of 2013-15)

Loading from Brownlow votes 2013-15: 65% for 65 votes

Patrick Dangerfield Value = [career length * career resources remaining]*[games per season]*[value of games]*10

=[(32.4-19)*60%]*[21.7]*[1.65]*10

=2859 points

We’re not suggesting that Dangerfield has another 286 games in him. What we are suggesting is that he’s one of the best handful of players in the competition, he can be expected to retire later than a standard player, and his games are much more valuable than the standard player. Meaning his projected remaining games may be worth as much as 286 by the standard player.

While quality is of course subjective, by constraining ourselves to seeing players’ games as maxing out at worth a bit less than double the average, we feel we’ve struck a balance between applying a high premium for quality while recognising the fundamental maxim that you need 22 blokes to make a team.

  1. Young players and recently emerged players

It should be apparent by now that with the focus on extrapolating future value from recent performance leaves a big gap for valuing young players without long track records, or players who have recently emerged as improvers. We discussed this with Cam McCarthy yesterday, and also several times last year.

Basically, our approach to valuing young players is to carry over their draft pick and to discount it for lost career time. Last year we did that linearly, deducting 1/12th of a player’s worth for every year older they got. That’s not ideal, as players don’t provide much of their value to clubs in their first couple of seasons and it can be argued that players lose very little value at all as new draftees just because they’ve aged a bit. We were certainly deducting way too much value from very young players.

Luckily, with the career resources calculations we have done based on age cohorts, we now have a discount rate that kicks in more slowly on draftees. We mentioned this look at career resources earlier, here’s the chart of outputs by age group and then the career resources chart again:

Draftees aren’t expected to have contributeed much. Compared to peak age cohorts around 27 years of age, 19-yearolds only produce 45% as much on average. Effectively, after two years, rather than having lost 16% of their value (2 of 12 seasons), young players still retain 90% (91% if rated elite via a top 5 Rising Star finish) of their value. Even after 4 seasons, rather than 66% they retain 78% or 80% of their value.

This is a discount rate we’re happy with, one that better reflects that a player’s prime years are in the second half of their 20s, but also reflects lost time. Remember that many of the very best, most valuable players in the competition (Fyfe, Dangerfield, Wingard, Hannebery, Dangerfield, Gaff, Cameron, Cotchin) started pumping out quality games almost immediately, so it’s not as though draftees lose no relative value at all after a couple of seasons.

A valuation issue remains for us with players who have significantly stepped up their output in a short space of time. The player who goes from fringe to playing every game (like Levi Greenwood last year or Callum Sinclair this year) is hard to project in terms of future worth because we don’t know that their output is sustainable. Our standard formula takes the last 3 years and projects future value based on that, because 3 years should provide a good sample in terms of selectability and durability. Such a 3 yearly snapshot tends to value recently emerged players very lowly because if you play 3, 2 and 22 games in successive years, that works out to 9 per season .

However, when we discuss such players we have and will continue note this issue, and will usually give a valuation based on their shorter term output as an indicator of how clubs might be evaluating them instead.

  1. Final thoughts

We don’t pretend our analysis is gospel. It’s built on the best data based measures and methods we can find or invent – a common currency for picks and players by comparing future games output, average retirement ages, loadings for elite quality, a discount curve based on age cohort outputs, an evidence-based valuation of draft picks – but in the end, it’s all only a starting point.

In the real world of trade period, clubs have different needs and priorities, clubs have different bargaining power in different trade scenarios, players have lots of intangibles beyond a numeric evaluation of their output, and all these things impact what clubs actually see as value and worth for players and picks.

However, what we hope to provide is a sense of a baseline, one look at what a “fair value” might be, that enables like-for-like comparison among picks and players. We hope this helps to inform discussions and extend analysis a bit further, we hope it can make clearer what sort of valuations and assumptions clubs are making, and where they are taking risks or being conservative.

Basically, we hope to use this sort of analysis to support and enhance discussion rather than to replace it.

8 comments

  1. Hey guys,

    One quick comment re: your valuation methods, given that Brownlow votes seem to disadvantage defenders (particularly key spoiling defenders), have you considered replacing Brownlow votes with internal club B+F votes? It may give a better indicator of how a club internally values a player.

    May help with better valuing players that are considered verging on elite but don’t draw many/any Brownlow votes (the Talia from Adelaide, Bashar Houli and Heath Shaw spring to mind)

    Cheers,

    1. We’ve gone some way to accounting for defenders with AA being incorporated – Shaw and Talia both rate a 20% loading based on that – but I agree that the problem of rating defenders is a tricky one. We just don’t have the data for it anywhere in our sport, and I think that is reflected in the poor showing of defenders in player of the year type awards.

      No reason we couldn’t factor high B&F showings in if we had the data, though that’s a lot to collect and then enter.

  2. Very interesting analysis.

    I do have a couple of points, however.

    Firstly, although I can see where you are coming from with your valuation of draft picks (and overall, I think it is very clever), it is worth noting that pick #1 very rarely turns out to be the best player from their draft year. It is certainly true that it is quite rare for a #1 pick to turn out to be a bad or mediocre player, but as they rarely turn out to be the best, assuming that they will have the highest level of future output seems flawed.

    Secondly, given that your All Australian/Brownlow vote eliteness modifier works out about the same as a player’s average Supercoach score, why not just use Supercoach scores?

    Finally, I think a lot of clubs would actually go for the GAJ for a 22 year old deal. As you pointed out, a player’s value is relative to their team rather than themself, so Ablett could be very attractive to a club vying for a premiership in the next year or two. And any bottoms clubs that employed crappy, short-sighted list managers may also go for it.

    1. Hi Jack,

      We actually agree with you regarding the value of high draft picks – number 1 often doesn’t turn out the best. It does, however, produce the highest average number of games. That’s the output we are mapping with the draft pick value chart. Your number 1 pick is the safest pick, so worth the most.

      Regarding Supercoach scores, we thought about it, but there’s two issues. The first is simple workload in terms of collecting the data and matching name labels. Given that we are heavily driven by quantity of output of total games played, we just wanted a pretty simple measure of who is producing the more valuable games. Supercoach does provides a measure of game influence but for now we’re pretty satisfied in what the 1% loading per Brownlow vote is providing us. The second issue is that while Bronwlow votes and AA has some flaws, Supercoach also does, particularly in undervaluing defenders and probably overvaluing ruckmen.

      Finally, I’m not sure many clubs would hypothetically trade for a 31 year old star any more given those guys can move as free agents if they get out of contract. One of the arguments for that free agency is really that clubs are reluctant to trade for old blokes because there’s not much value in them. Not saying nobody would, ujust that it’s likely to be a terrible deal when looked at in terms of total future returns. Remembering again that we’re not claiming an all encompassing assessment of value, just using a method to compare values of picks and players based on quantifiable future yields, in order to see where the risk and upside lies. In that sense, getting 20 games out of GAJ in exchange for a decent kid is a huge risk. It pays off if you win a flag that one year, but otherwise you’ve turfed a bloke who could serve you for ten years. We would rate the trade as massively lopsided, and what that would be illustrating is where the balance of risks lies – the club taking the old guy is betting hard on immediate payoff and it’s not as likely to work out as the 22 year old is. As ever, the numbers serve the analysis and commentary, they don’t replace it.

      1. Hi Sean,

        I don’t mean to say that the GAJ trade would necessarily be a good one; merely that I think some clubs would (sensibly or otherwise) go for it.

  3. Love your work, but I don’t understand why you treat “games played” as something that players produce and clubs want (e.g. “at most [a Number 1 pick] can provide 22 of the 484 games needed by a team”), as if clubs are seeking to acquire as many “games played” from players as possible. Surely it’s the other way around: a club has 22 “games played” it can spend each week, and it wants to extract maximum return from the players it spends them on. That is, “games played” is a scarce resource owned by the club.

    For example, clubs need to put games into young draftees, during which time everyone knows their output will be below their potential – and not only that, but also below the output of an older but more limited “list clogger” type player. That is, these “games played” by the draftee are actually hurtful to the club in the short term. In an ideal world (for the club), it would acquire top talent that had already played 50 or 100 games!

    Also, all clubs have several players that are on the fringes of team selection. It may pick Player A over Player B, but their output will be very similar (and easily replaced). In this situation, it’s clearly wrong to say that Player A represents much more value for the club than Player B (according to “games played”), since they can be easily swapped with no decrease in team performance.

    Doesn’t it make more sense to consider “games played” as something a club needs to spend rather than wants to acquire?

    1. Hi Max,

      A random poster here (I don’t represent the author or the blog). I’m also not an economist (as I gather you are) so it’s possible that I’ve not understood your question fully – my apologies if that’s the case.

      As I understand it, games played is a proxy for the player’s quality and (therefore) an indicator of future output. i.e. a 23 y/o who has played 85 games is of higher quality than a 23 y/o who has played 15 games. As a proxy for a player quality, it represents a scarce resource held by a player’s present club in any trade negotiations.

      Also, your comment seems to imply that the trading parties are clubs and players. But here, the trading parties are clubs and clubs. The question of the player’s ‘fair value’ goes to what Club A should give Club B for player X. It’d be interesting to see a model for a player’s ‘fair value’ in either dollar terms or in terms of % of salary cap, but that’s a separate exercise to the one HPN is carrying out here.

  4. Hi HPN

    Quick question on the “Career resources remaining” tables shown under points 3 and 6 on this page: to what extent in your model for resources remaining relates to quantity rather than quality of output? (or, alternatively, what you would estimate this balance is) By quantity I mean (for example) pure possessions rather than, say, kick accuracy

    Enjoying the website!

    Cheers

Leave a Reply to Sean Cancel reply

Your email address will not be published. Required fields are marked *