Trade Value vs. ADP: Building a Trade Evaluation Formula

Alex Eifler

spiller

Editor’s Note: This article is penned by a new Member Corner writer, Alex Eifler We welcome Alex to the crew and look forward to his contributions moving forward. This is the first in a series on Trade Value vs. ADP and an attempt at building a trade evaluation formula.

Introduction

This article is intended to be the first part in a series where I attempt to bring together some of the outstanding work done by other DLF writers and marry it to the trade review threads posted in the forum. The eventual goal will be to identify discrepancies between ADP and the trade markets, which could be exploited by savvy dynasty owners.

This first article is dedicated to defining the value of the various dynasty assets (veterans, rookie picks and open roster spots). It then uses the resulting model to assess five recent trades from the forum, with an evaluation of how well the model fits the forum consensus.

All feedback on the model outlined below is appreciated. I look forward to fine-tuning it prior to the next part in this series, where I hope to evaluate a far higher number of trades.

Before I begin, we should talk about general assumptions. I will use rules and scoring consistent with Ryan McDowell’s monthly startup drafts, since I used his ADP data (specifically the February ADP data, which was the most current data available). I will assume all trades use these same rules and scoring, and avoid analyzing trades where there are major departures (e.g., 2QB, superflex, IDP, TD only, devy).

Assigning Values to Dynasty Assets

Defining the values of all available dynasty assets is the first step in building the model.  I will go through how I assigned values for veterans, rookie picks, and open roster spots in the paragraphs that follow. I have cited my sources whenever possible, since they are the ones who did the hard work. I am standing on the shoulders of giants and I can confirm that Jimmy Johnson’s hair looks spectacular from up here.

Veterans

The dynasty community has recently been using startup ADP as a useful substitute for value. While I believe there are some limitations associated with this reasoning, I think it is a great starting point for defining the value of a veteran player.

Using the February Startup ADP as a baseline, the next step was to figure out how to assign value to each player. The first thing that came to mind was Jimmy Johnson’s Draft Value Chart, which was designed to value the draft slots in the NFL Draft. Valuing NFL draft picks and valuing players in a dynasty setting have a lot in common, since both are long-term investments in players of various levels of ability, so I started there. The Johnson Chart (no jokes, please) has been criticized over time, primarily for overvaluing the top picks and undervaluing the mid- to-late round picks. I set out to see what else was available.

I came across the work of Kevin Meers at the Harvard Sports Analysis Collective, who has used Pro Football Reverence’s Career Approximate Value to create a more realistic approximation of what can be expected from the career of the player selected in each draft slot. The only issue with using this chart for dynasty players instead of NFL draft picks is that there is a 25-point drop off between picks 32 and 33, corresponding to the end of the 1st round and beginning of the 2nd round in the NFL draft. Since this doesn’t make sense in valuing dynasty players, I tried a few methods of adjusting the values. First, I tried increasing the value of all picks after 32, and smoothing the decline in value between picks 33 and 43. The issue with this approach was that late-ADP players were too valuable relative to the top players. The approach I settled on was to smooth the values between pick 25 and pick 50, but leaving the values of picks 1-24 and 51+ as they were. I prefer this approach because it maintains the low value of late picks from the Harvard Chart, while eliminating the sudden drop. I call my chart the Amended Harvard Chart.

Using this chart as a starting point, I performed a lookup for each player’s ADP on the Amended Harvard Chart, rounding ADP up when it was fractional (5.7 rounds to 5). For example, the top player (A.J. Green) has an ADP of 1.8. This gets rounded up to 1, and his value is the value in the Amended Harvard Chart corresponding to the first pick (494.6). Adrian Peterson (ADP 22.2) and Brandon Marshall (ADP 22.5) both would both receive the value of the 22nd pick, 232.1. Since the Harvard Chart has only 224 picks, but the ADP list has 241 ADP slots, I simply continued the linear pattern of the seventhround on the Amended Harvard Chart through pick 241 by subtracting 0.4 from the value of each subsequent pick).

Rookie Picks

Rookie picks play a key role in many dynasty trades, so properly valuing them is going to be critical. Building on the work Rob Pisano did in his recent Appraisal of Rookie Draft Picks, I used each rookie’s position in the February startup ADP data to generate the value of each rookie pick. For example, since Sammy Watkins (ADP 29.5) was the first rookie selected in terms of ADP, he stood in for the 1.01 pick; Jace Amaro (ADP 103.2), as the 12th rookie selected, stood in for the 1.12 pick. Using this methodology, I assigned all the rookie picks through 4.11 (Zach Mettenberger, ADP 240.5).

Values were then attached to each pick in the same way as values were assigned to the veterans, using the Adjusted Harvard Chart value for that ADP using a lookup.

I suspect rookie values fluctuate substantially leading up to and after the NFL draft, so please consider these values to be as of February 2014.

If anyone has an idea for how to value future year draft picks, please let me know in the comments. I suspect the answer lies in the intersection between Zach Bahner’s recent article and this one, but I need more time to think on this.

Open Roster Spots

An often-ignored aspect of dynasty trades is the value created by freeing up a roster spot. This is accomplished by trading two-for-one (or three-for-two, etc.) and allows the team giving up more players to pick up a player off waivers to fill the vacancy.

My first instinct for valuing an open roster spot was to peg the value to the best available players on waivers. Obviously, the best available player on waivers varies from league to league, and depends greatly on the size of your league, number of roster spots, and starting requirements. I tried an approximation of 40 (ADP 200 or so), since this is in the range of some perpetual waiver-wire players (Tim Wright, Griff Whalen, Benny Cunningham, Joseph Randle, Andre Holmes and Rob Housler). Even if most of these players are owned in your league, chances are that at least one player in this range is available. If your league is exceptionally deep (26+ player rosters), you may want to use a lower value.

Comparing this value to the rookie pick values, an open roster spot is worth roughly pick 3.11. This seems correct on the face, since fourth round picks have very little value in most dynasty leagues. If you find that your league-mates consistently use their fourth round picks, you again may have to adjust the value of open roster spots down from 40.

While it is obvious that an open roster spot should be included in the equation when there is a two-for-one trade, it is less obvious whether to include one when draft picks are involved. In the case of draft picks, the proximity of the rookie draft plays a part. For example, it is clear that, in a draft-day trade of 1.05 and 1.06 for 1.02, there should be an open roster spot given to the owner trading away the two picks, since he will have the flexibility to keep one of his current players or pick up a player from waivers. It is less clear that the open roster spot is as tangible or valuable when picks are traded mid-season, or when future draft picks are traded.

For simplicity’s sake, I have assigned the same value to all open roster spots, and have included an open roster spot in any trade where an unequal number of assets change hands. I would consider diluting the value of open roster spots created by draft pick trades, depending on the amount of time until the rookie draft though. Let me know in the comments what you think.

Evaluating Trades: Does the Model have Construct Validity?

When evaluating an argument or statistical model, one of the critical testing points is whether or not there is construct validity. Construct validity refers to the degree to which a test measures what it claims, or purports, to be measuring (thanks Wikipedia). In this case, we can test construct validity using the trade threads that are available on the forum. In a few very useful cases, trades go through a series of iterations, where one side proposes a trade and the other side counters. If the values that the model produces match directionally with the consensus of the forum contributors, it means the model is consistent with human trade evaluation. For the purposes of our model, that would be a good thing.

Trade 1: Blockbuster trade, with many draft picks changing hands

I will start with the trade featured in this thread, since it is a good test for the model and there were forum contributors on both sides of the deal, with a number saying that the deal was fair.

trade1

The model agrees the deal is fairly even, since the difference between the sides (46.6) is roughly equal to a mid-3rd round pick. There is a slight advantage for Team B, which corresponds to there being a few more forum voices on the Team B side. So far, the model is doing well.

Trade 2: Jamal Charles vs Gronk and 1.01

The next deal to review is emblematic of many dynasty deals, where a package of a player and a pick is used to acquire a stud.

trade2

Once again, the model determines the deal to be quite fair, with a difference in value of only 15.3.  The forum agrees with this assessment, though a few contributors came down on the Gronk side.

Trade 3: Two studs vs. a bunch of picks

I picked this trade because I expected it to be difficult for the model, since your opinion of the trade is based largely on your personal style as a dynasty owner. Owners who like potential will side with Team B, while owners who like a sure thing will side with Team A. The inclusion of conditional picks is also problematic.

trade3

Note: Team A also gets conditional 2015 1st and 2016 2nd round picks, if Julio and Marshawn fail to have a combined point total of 382.5 by the end of 2014. I was not able to assign a value to the conditional picks, so I left them out of the equation. Also, this league has 14 owners, so pick 1.14 was converted to 2.02, and 2.04 was converted to 2.06.

This thread featured a vote, which was a decisive 33 to 7 victory for Team A. The model has this trade as exactly fair, with difference of only 0.8. The conditional draft picks throw a wrench in the works, effectively offering protection for Julio Jones and Marshawn Lynch. If either of them is substantially injured or ineffective in 2014, the trade becomes lopsided in Team A’s favor.

Because our model is not yet able to value future draft picks or conditional situations, the model doesn’t work well here. We can only say that, if Julio and Marshawn are both healthy and productive, this is an even trade.

There is also the question of whether there are diminishing returns when you own so many high rookie picks in a single draft.

Trade 4: Two trades

It is helpful when multiple trades are evaluated by the same forum contributors, since we get double the data points. This series included a large trade and a smaller trade, with Team B being the same team in both trades.

trade4

The forum consensus was a strong win for Team B on the first trade, and a slight win on the second trade. This is exactly what the model shows.

Trade 5: Offers and Counter-Offers

This last evaluation is not a trade per se, but a series of offers and counter-offers. Our goal would be for the difference in value between the two sides to narrow as the process goes along, and the deal gets closer to being accepted. Therefore, I have included a difference column for this series.

trade5
There was near total consensus that the first and second offers were lopsided in favor of Team A, and the offers ware rejected by Team B. The third offer was a counter-offer from Team B, which was rejected by Team A without much discussion transpiring on the forum. The fourth offer was on the table at the time this article was being written, and there was no consensus as to which side was getting more value.

The model tracks well to the forum consensus, with more value going to the side that made the offer in each case. The difference in value between the two sides narrowed as the deal came closer to being accepted. The final offer is almost exactly even. 

Conclusion: did we build a trade evaluation formula?

Our model appears to be roughly in-line with the consensus on the forum, and the size of the value difference corresponds to the strength of the opinions expressed. In that sense, our model has construct validity and can be reliably used to determine which side is getting more value in a trade.

There are obviously some limitations of the model, listed here in rough order of gravity:

  1. Unable to account for roster/lineup needs or league-specific rules
  2. Limited to the baseline ADP data that is available, so it only applies for PPR leagues with standard lineups
  3. Only as good as the ADP data that comes out of the DLF Mocks
  4. Unable to value future draft picks and conditional picks (I will attempt to correct this, perhaps using Zach Bahner’s recent article as a guide)
  5. Should be considered applicable to the current point in time, as player and pick values swing wildly around the NFL draft

The underlying assumption in the construction of the model – that the Adjusted Harvard Chart values are appropriate for use with dynasty players – appears to be plausible, given the model’s performance in evaluating trades.

As I said before, I am very open to feedback on how to improve this model, and would appreciate any ideas on how to value future draft picks. Please let me know in the comments if you have any ideas.

My next goal will be fine-tuning the model through feedback and trade analysis, and possibly creating a simple tool that is publicly available. After that, I will begin looking for systematic discrepancies between ADP and the trade markets, which could be exploited by savvy dynasty owners.