The National Football League (NFL) draft has been held since 1936. In the inaugural draft, only 24 players out of a selected 81 chose to continue their football careers. The majority, including the number one overall pick and 1935 Heisman Trophy winner, John Jacob “Jay” Berwanger, decided that there was no money in pro football. He chose instead to become a foam-rubber salesman ([3]).

Today, the draft consists of seven rounds (the length of the draft since 1994). Under the NFL’s reverse-order-of-finish draft, the weakest teams draft first and the winner of the Super Bowl drafts last. A team may trade a draft pick in a given round for one or more additional lower picks in the same or later rounds or for one or more picks in future drafts (or a combination of the two).

In the early 1990s, Mike McCoy, a Dallas Cowboys vice president, and then Cowboys head coach Jimmy Johnson developed a chart to help evaluate the value of draft picks (https://www.drafttek.com/NFL-Trade-Value-Chart.asp). In their research on market efficiency in the NFL, [2] conclude that the top NFL draft picks are overvalued by Jimmy Johnson. They use data on draft-day trades for the years 1988 through 2004 to develop a model for predicting the market value of draft picks relative to the overall number one pick. Their analysis of player performance measures (such as the number of games started, the number of games played, and the probability of making the Pro Bowl) leads them to conclude that teams put “too high a value of choosing early in the draft”.

The paper of [4] on winners and losers in the NFL draft endeavors to value each draft pick based on how all players drafted between 2002 and 2014 ultimately performed in their careers based on their approximate value (hereafter AV), a comprehensive metric reported for each NFL player at Pro-Football-Reference.com. That is, Sommers relates each draft pick number (1 through 224) to the average career AVs of all players drafted by the NFL between 2002 and 2014. Sommers then uses career AVs to come up with a different value chart that reflects the Massey and Thaler contention that players near the top of the draft (in early rounds) are overvalued relative to the Johnson value chart. [1] use Round 1 NFL trades on Day 1 (hereafter, “Day 1 trades”) between 1985 and 2005 for 359 players (97 percent of whom had completed their careers by 2015) to assess which value chart—Johnson or Sommers—did a better job of predicting which team’s picks in a Day 1 trade would end their careers with the higher AV.

This paper updates the Jung and Sommers comparison to include all Day 1 NFL draft trades between 2000 and 2010. All players involved in these trades are no longer playing in the NFL. Hence, their “career” AVs are not subject to change. Does the Sommers point value chart outperform the Johnson point value chart in predicting winners versus losers in Day 1 NFL draft trades?

The Jimmy Johnson trade-value draft chart is reproduced in Table 1. For each of the 224 picks (seven rounds, with each of 32 teams picking once per round), the Johnson chart assigns a point value to each draft pick. The first overall pick is worth 3000 points, the second is worth 2600 points, and so forth until the last or 224^th pick—worth only 2 points.

The point allocations are front-heavy, with the entirety of the 7^th round being assigned a cumulative point value less than that of the third pick in the 3^rd round and less than 9 percent of the 1^st overall pick. The Johnson chart nonetheless enjoys widespread use to help NFL teams approximate the market value of draft selection swaps.

The Sommers trade value chart is reproduced in Table 2.

While the last (or 224^th) pick on Johnson’s value chart receives only 2 points, Sommers’ chart awards the last pick 298 points—on the same scale where the first

Table 1. Jimmy Johnson NFL draft trade value chart.

Source: http://www.pro-football-reference.com/draft/draft_trade_value.htm.

Table 2. Sommers NFL draft trade value chart.

Source: [4].

(or number 1) pick receives 3000 points. The drop in value using the Johnson chart from the first pick to the last pick in Round 1 alone is over 80 percent (viz., 3000 to 590). By comparison, the drop in value using the Sommers chart from the first pick to the last pick in Round 1 is less than 58 percent (viz., 3000 to 1270). The Sommers chart places less emphasis on the early round picks and more accurately values late-round sleeper picks. Perhaps the most famous “sleeper pick” is quarterback Tom Brady. Taken in the 6^th round, Brady went on to win seven Super Bowls, voted MVP in five of them, appeared in 15 Pro Bowls, and was declared league MVP three times. Brady’s 199^th pick is allotted just 11.8 points in Johnson’s chart versus 357 points in Sommers’ chart.

All of the trades on Day 1 of the NFL draft between 2000 and 2010 are from the Pro Sports Transactions Archive (https://www.prosportstransactions.com/football/DraftTrades/Years/). Over this 11-year period we recorded 58 draft-day trades involving draft picks from only the current year (47) and future years (11). The team with the first round pick who is designated in prosportstransactions.com as the “Player Drafted” is Team A; the team that acquires the “Player Drafted” and possibly other draft picks from Team A is Team B. We exclude trades that involve players already in the NFL. All 58 draft-day Round 1 trades involve only players with a round and pick number. That is, there are no players outside the 224-pick chart in our final sample. For each team involved in a trade, we recorded the acquired player’s name, his overall pick number, and his chart value according to Johnson and Sommers. For each drafted player (and other draft picks involved in a trade), we also collected data from Pro-Football-Reference.com on the drafted player’s weighted career AV (hereafter wAV), a weighted sum of a player’s annual AV scores, where the weights start at 100% for their best season, 95% for their second-best season, 90% for their third-best season, and so forth. (The wAV should not be confused with “career AV” which is just the unweighted sum of a player’s AV scores.) The chart values for future picks in an ex ante expected slot were discounted at a 5 percent rate, about equal to the 3-month Treasury Bill rate in January 2007 (4.98 percent) before it plummeted to 0.05 percent in December 2009. For example, the Johnson chart value of the overall 8^th pick next year would be 1400/1.05 or 1333.33 points. Players may be counted in more than one trade. For example, Dez Bryant was drafted 24^th overall in the 2010 draft, but his pick was involved in three trades, passing the pick from the Philadelphia Eagles to the Denver Broncos, New England Patriots, and finally to the Dallas Cowboys. His pick is eligible to be counted in each trade because each team acquired the same 24^th pick. We added up each team’s point valuations. Finally, for each team’s acquired picks, we added up the players’ NFL career wAVs. The data on all 58 trades appear in Table 3. On closer examination of Table 3, the Johnson chart predicts correctly which team will have the higher wAV in 31 of the 58 (53.4 percent) of the trades. The Sommers chart predicts correctly in 42 of the 58 (72.4 percent) of the trades.

Table 3. NFL draft trades and trade values, Day 1, 2000-2010.

For each trade and each trade value chart—Johnson or Sommers—we derived a disparity measure, D. If the Johnson chart, for example, predicts that Team A has more point value than Team B, then the disparity value is equal to the cumulative career wAV of all picks acquired by Team A minusthe cumulative career wAV of all picks acquired by Team B. That is, if the Johnson chart correctly predicts the i^thtrade, then the disparity measure, D_i, will be positive. If, however, the Johnson chart predicts one team has a point advantage, but in fact the other team ends up acquiring players with a higher cumulative total of wAVs, then the disparity measure will be negative. We investigate whether the average value of D_Johnson(or D_Sommers) is equal to zero against the one-tailed alternative hypothesis that the average value is greater than zero. If we cannot reject the null hypothesis that the average value of D_Johnson(or D_Sommers) is equal to zero, then the chart’s ability to distinguish good trades from bad ones is no better than a coin flip.

For each chart, we regress the difference between the total career wAVs of Team A’s draft picks minusthe total career wAVs of Team B’s draft picks (career_wAV_difference) against the corresponding difference in the total point values of draft picks (Johnson_point_differenceor Sommers_point_difference) as follows:

career_wAV_difference= β₀ + β₁point_difference+ ε_i (1)

where ε_i are independent errors. If the estimated slope coefficient, b₁, is not discernible from zero, then variation in the difference between Team A’s point values less Team B’s point values tells us nothing about the variation in the team differences in career wAVs. If, however, the estimated slope coefficient is significantly greater than zero and if the difference between the point values of Team A’s picks less than the point values of Team B’s picks is positive (negative), then the difference between career wAVs of drafted players acquired by Team A less career wAVs of those players acquired by Team B will also be positive (negative).

Which chart—Johnson or Sommers—better predicts career wAVs, in other words, good trades from bad trades?

For the 58 draft trades, the average Johnson chart disparity, D - _Johnson, is 38.36 with a one-tailed p-value of 0.070. In other words, the average disparity using the Johnson chart is not discernible from zero (using a 0.05 level of significance). The average Sommers chart disparity, D - _Sommers, is 777.72 with a one-tailed p-value less than 0.001. In other words, over the 11-year period, the Sommers chart was more often than not correct in predicting the two teams’ difference in weighted career AVs. That is, if the Sommers chart indicates that Team A acquired more points than Team B, then at the end of their careers players acquired by Team A will have significantly higher career wAVs than did the players acquired by Team B.

When the difference in Team A’s less Team B’s weighted career AVs was regressed against the difference in Team A’s less Team B’s Johnson point totals, the results were (standard errors in parentheses):

career_wAV_difference = 23.192 − 0.015 Johnson_point_difference (2)

(6.682) (0.034)

R² = 0.003

The corresponding regression results using the Sommers point totals were:

career_wAV_difference= −7.969 + 0.038 Sommers_point_difference (3)

(12.504) (0.014)

R² = 0.119

Using the Johnson chart, Equation (2) indicates that knowledge of the difference in point totals ascribed to the draft picks of Teams A and B sheds no light on the ultimate weighted career AVs of players drafted by the two teams.

Figure 1.A scatter plot relating the weighted career AV difference to the Johnson point difference for each of 58 Day 1 NFL draft trades, 2000-2010.

Figure 1 underscores the anemic predictive value of the Johnson chart. The best-fitting line through the scatter of points is negativelysloped, although this slope is statistically no different from zero. Over 70 percent of the Johnson point differences are within 100 points of zero. And, the cluster of points around a point difference of zero (measured along the horizontal axis) explains why the Johnson chart cannot explain any variation in weighted career AVs. Because NFL teams followed the Johnson NFL draft chart so closely (between the years 2000 and 2010), there is not enough variation in Johnson point differences to explain any variation in players’ future weighted career AVs.

Equation (3) indicates that if the point totals of players drafted by Team A exceeds that of players drafted by Team B, then the ultimate total weighted career AVs of the players acquired by Team A will exceedthe corresponding total weighted career AVs of the players acquired by Team B. In other words, the slope coefficient in Equation (2) is not discernably different from zero (p = 0.667); the slope coefficient in Equation (3) is positive and significantly different from zero (p = 0.008).

Figure 2 shows that the best-fitting line through the scatter of points is positively sloped. Larger differences in Sommers point totals leads to larger differences in future weighted career AVs.

Eleven of the 58 Day 1 trades involved chained trades where the same Round 1 pick appeared more than once: Shaun Ellis (Pick 12, hereafter P12, in 2000), T.J. Duckett (P18 in 2002), Ty Warren (P13 in 2003), Chris Gamble (P28 in 2004), Tye Hill (P15 in 2006), Duane Brown (P26 in 2008), Jeremy Maclin (P19 in 2009), Clay Matthews (P26 in 2009), Brandon Graham (P13 in 2010), and Dez Bryant (P24 twice in 2010). When multiple observations on these Round 1 picks were collapsed to one observation per pick chain (that is, the first time the Round 1 pick is involved in a Day 1 trade), Equations (2) and (3) were re-estimated. The slope coefficient in Johnson Equation (2) is −0.006 (standard error = 0.038), R² = 0.0005 and the slope coefficient in Sommers Equation (3) is 0.046 (standard error = 0.017), R² = 0.145. That is, the main results hold.

Figure 2. A scatter plot relating the weighted career AV difference to the Sommers point difference for each of 58 Day 1 NFL draft trades, 2000-2010.

An NFL draft valuation chart that can consistently predict draft pick trade outcomes in terms of actual (weighted) career AV would be highly valuable to general managers and coaching staffs of NFL teams.

The Johnson chart has enjoyed widespread use. But, if players in the early rounds of the NFL draft are overvalued relative to players in late rounds, then an alternate chart that assigns more value to players drafted in, say, rounds 4 through 7 may align better with the weighted career AVs of these late-round drafted players.

The results presented here show that the Sommers chart values are not only better at predicting the “winner” of a draft trade, but that over an 11-year period (2000 to 2010) there is a strong relationship between the Sommers point difference (Team A’s total minus Team B’s total) and the actual career weighted AV difference of these acquired draft picks (again, Team A’s total minus Team B’s total). The better predictive ability of the Sommers point difference is largely because when the Johnson trade-value chart is used (which appears to be quite often), the total point values of the draft picks of the two teams are roughly the same. Many NFL teams that use the Johnson chart believe they have consummated a fair trade of draft picks. The Sommers chart that relates draft pick numbers to career AVs suggests otherwise.

The foregoing analysis covers only Round 1 Day 1 trades. Moreover, the Round 1 Day 1 trades discussed in this paper exclude trades involving veteran NFL players or trades involving compensatory picks outside the 224-pick chart.