*This is the first post in HPN’s coverage of the AFL off-season. For up-to-date information on the latest draft order and list information, we recommend Draft Guru.*

A couple of years ago HPN unveiled its first Draft Pick Trade Value Chart (TVC). We felt, considering the amount of the time that has passed since we thought we would build on what we had before.

Without further ado:

Here are the raw values against the adjusted values (more on the process below):

Before you read any further we suggest you read up on this post we did last year on the theory of valuing trades, draft picks and players. Warning, it’s a long read.

### What has changed

In order to align it with our player value trade formula, we have decided to weight the game outputs per pick by Brownlow Medal votes per pick. This has created a universal value across both picks and players – both picks and players have a standard value (games played/projected) and an elite loading for more valuable games (Brownlow Medal votes received/projected).

This is a visual representation of how the first and new versions of the charts compare:

In the past we had used a rough compounding of the games played formula to estimate the elite loading (that is, we manually upped the top few picks a bit), but upon testing the data again recently we thought we’d take a proper look at the chart.

The result is that, as per standard perception, that the top picks have become more valuable. The general profile of the chart remains the same, but the top end rightfully gains more value.

The last two years have also provided a more full accounting of the existing dataset with more careers coming to completion.

### What hasn’t changed

The untapped value of the second round is still apparent on our charts, with only 10 games of adjusted output the difference between the 19th and 36th pick. The decline in the first round remains steep, and it also increases in the third round, but the second round really does seem to be a good value place for obtaining draft picks. Essentially, draft order in the second round seems barely to matter and that suggests that two-for-one trades of picks down the order in the second round would be a smart move for a club.

Our Player Trade Value Formula (PTVF) remains the same for now as well, surviving the microscope for now.

We have been working tying the elite output to other multifactor outputs, but preliminary tests have found that Brownlow Medal votes to be a decent (and simple) proxy for fantasy point loading. Expect to see the results of this either later this trade period or next year.

### Data Notes

*Wonk/nerd alert. Don’t say you weren’t warned.*

All the data we used in creating the HPN TVC was sourced with permission from Draft Guru.

There are 993 picks in the dataset. The make up of the dataset is explained below.

The first year of our sample is 1993, as it is the first year of a true open draft. Some may point to the 1988 as the first draft, others to 1981. Whilst there were earlier drafts than 1993, they were all heavily geographically restricted, and largely operated whilst the zone and u18 system still was operational. 1993 was the first season where two Victorian players were chosen in the top 3, and national representation became a wider concern in the draft. If someone wants to argue us around, we’re more than happy to hear them out.

We’ve deleted all Father/Son picks through this era, as they were locked to the third round, and don’t necessarily represent fair value. When we begin to consider data from the bidding era, we intend to add the Father/Son and Academy picks from the nomination and bidding period, as the effect on the data should be minimal. This impacted 33 selections, or less than 5% of the total sample. We did not move the next pick up the order, but instead relied on the use of average games for each pick to keep the formula consistent.

We also deleted two picks that were forfeited during this period due to salary cap infringements (Essendon’s pick 40 in 1996 and West Coast’s pick 42 in 1998). We did not move the next pick up in the order, but instead relied on the use of average games for each pick to keep the formula consistent.

Passed picks *do* contribute to the usage values of each pick, as it is an adjudication that *no-one *would be a better use to the club than an actual player. While some were held to be taken in the Preseason draft, each pick in that draft had to occur *after* the last selection in the main draft, creating an implicit value judgement regarding that player and that national draft pick in the decision.

We have cut off our sample with 2004, for a couple of reasons. Firstly, it is the cutoff we used before, and how we can judge what difference made by the new elite weighting. Secondly, the HPN retirement age formula has a standard retirement age of 31 years old, placing the remaining 2004 draft players only 2-3 years away from retirement on average. Thirdly, there are only 12 players who were drafted in 2004 still on AFL lists, while there are 17 players from 2005 still on lists. Although the difference seems minor, the continuing cohort of pre-2004 players isn’t much more than that 17. We did some testing of the data, imputing the expected career output of these players and the potential impact on the chart, and even if the remaining players all went on to break Boomer Harvey’s games record, the impact on the chart values would be minimal due to the polynomial regression applied. We noted that including the 2005 draft onwards begins to shift pick values downward significantly, and as such we have excluded them for this version.

The polynomial regression we used was a polynomial sixth degree. We did this primarily for reasons of best fit, and simple common sense. We noted that whenever a higher degree was tested, there were occasionally picks that were lower yet had a higher value. The exact formula was:

Polynomial degree 6, 89 x,y data pairs.

Correlation coefficient (r^2) = 0.7244902102537171

Standard error = 319.42823720380517

Coefficient output form: mathematical function:f(x) = 2.9878013738618615e+003 * x^0

+ -2.9874235116293801e+002 * x^1

+ 1.8167034931156536e+001 * x^2

+ -5.6415161528188662e-001 * x^3

+ 9.1802650122308205e-003 * x^4

+ -7.5097333757060268e-005 * x^5

+ 2.4322052268311531e-007 * x^6

Our previous formula had a r^2 of 0.7481, and the current AFL Draft Pick Value Chart has an r^ of 0.7482 (albeit using a logarithmic regression and a different dataset). We noticed that for every year of data we introduced to the dataset, the r^2 increased. For example, stretching it to 2006 saw the r^2 increase to 0.735, and for the entire dataset to 2015 saw an r^2 of 0.796. Obviously we couldn’t use one-year-old data, but we feel over time the method will improve in accuracy.

We feel that considering the information we had, that this chart is a massive step up on our last chart.

## 4 comments