Sie sind auf Seite 1von 17

THE LOSERS CURSE: OVERCONFIDENCE VS.

MARKET EFFICIENCY IN THE NATIONAL FOOTBALL LEAGUE DRAFT

SUPPLEMENTAL ANALYSIS

August 2011

Abstract

Supplemental information including methodological detail, descriptive data, additional results and figures.

Table of Contents Appendices S.1 Draft-pick estimation methodology for trades including future picks.

Tables S.1 S.2 S.3 S.4 Trade Frequency by NFL Team and Role in the Trade Player Costs and Benefits Spline Regressions and Surplus Quantiles Traded-for Players

Figures S.1 S.2 S.3 S.4 S.5 S.6 S.7 S.8 S.9 S.10 Chart Convergence Draftee Compensation, by Draft Order Memory Function in Performance Value Model Mean and Standard Deviation of Surplus Value, by Draft Pick Convexity of Performance Value Spline Regressions and Surplus Quantiles Surplus Value Calculated Using Only Offensive Lineman Surplus Value Calculated Using Only Wide Receivers Portfolio Analysis: Gains From Trading Down Portfolio Analysis: Gains From Trading Down (Compensation-constant)

Appendix S.1 Draft-pick estimation methodology for trades including future picks

In order to value future picks we modify (4) to include a discount rate, , giving us

(5)

e ( ti 1) , v(t ) = (1 + ) n
r

r i

for a draft pick n years in the future. This expression reduces to (4) when n=0, i.e., the pick is for the current year. Substituting (5) into (3) and solving in terms of the highest pick in the trade, which by definition is in the current year, gives us

(6)

1 j n e ( t L 1) H log t1 = j =1 (1 + ) n

e +1, i = 2 (1 + ) n
m ( tiH 1)

which simplifies to (5) if all picks are in the current draft.

Table S.1 Trade Frequency by NFL Team and Role in the Trade 1983-2008. Excludes trades involving future picks or players.
Franchise ARI ATL BAL BUF CAR CHI CIN CLE DAL DEN DET GB HOU IND JAC KC MIA MIN NE NO NYG NYJ OAK PHI PIT SD SEA SF STL TB TEN WAS Total Up 4 10 13 4 6 5 2 8 17 10 17 16 1 8 9 7 12 11 17 7 5 14 25 12 9 6 7 20 10 6 5 11 314 Down 7 9 8 5 7 17 5 3 21 11 6 25 4 7 10 2 17 8 18 9 3 9 8 9 6 4 12 15 9 15 14 11 314 Total 11 19 21 9 13 22 7 11 38 21 23 41 5 15 19 9 29 19 35 16 8 23 33 21 15 10 19 35 19 21 19 22 628

Table S.2 Player Costs and Benefits 1994-2008 seasons, in 2008 dollars. Includes all drafted players in their first five years in the league, excluding punters and kickers. Variable All Rounds Performance Value Compensation Surplus Value 1st Round only Performance Value Compensation Surplus Value 2nd Round only Performance Value Compensation Surplus Value N Mean SD p50 p90

17,155 1,703,390 2,083,081 17,155 1,044,029 1,437,752 17,155 659,361 1,626,597

941,893 4,776,152 606,073 2,548,188 251,318 2,983,386

2,292 4,097,102 2,968,115 3,610,202 8,987,901 2,292 3,080,305 2,130,686 2,477,002 5,445,122 2,292 1,016,797 2,809,072 462,634 4,339,132

2,310 2,534,196 2,059,345 1,942,033 5,207,619 2,310 1,362,362 1,271,921 1,067,390 2,450,663 2,310 1,171,834 1,907,769 762,785 3,861,268

Table S.3 Spline Regressions and Surplus Quantiles Regressing estimated surplus value (performance value compensation cost) on a linear spline of the draft order. The spline is knotted between rounds, so that the x variables reflect the estimated slope, i.e., the change in surplus, during a draft round. Models (2)-(4) are quantile regressions. Observations are player-season. Sample is limited to all drafted players, excluding punters and kickers. Reported in 2008 dollars. (1) Mean Round1 0.0248*** (0.0027) Round2 -0.0166*** (0.0022) Round3 -0.00812*** (0.0022) Round4 -0.00160 (0.0022) Round5 -0.00580*** (0.0022) Round6 -0.000312 (0.0021) Round7 -0.00194 (0.0014) Constant 0.598*** (0.061) Observations 17155 R-sq 0.03 Pseudo R-sq . (2) p25 0.0478*** (0.00045) -0.00186*** (0.00035) 0.000318 (0.00035) -0.000242 (0.00035) -0.00101*** (0.00036) 0.000473 (0.00034) 0.0000134 (0.00023) -1.532*** (0.010) 17155 . 0.02 (3) p50 0.0168*** (0.00000) -0.00333*** (0.00000) -0.0115*** (0.00000) -0.000544*** (0.00000) -0.00689*** (0.00000) -0.00126*** (0.00000) 0.00000 (0.00000) 0.199*** (0.00000) 17155 . 0.02 (4) p75 0.0122*** (0.0022) -0.0592*** (0.0018) -0.00293 (0.0018) -0.00260 (0.0018) -0.0121*** (0.0018) -0.00512*** (0.0017) -0.0000573 (0.0012) 2.702*** (0.050) 17155 . 0.09

Table S.4 Traded-for Players Performance of players as a function of draft pick and whether or not they were the target of a draft-day trade. Targets are defined as the player taken with the highest draft pick acquired by the team trading up. Performance is measured in a players first five years, in four ways: probability of being in the NFL, games played in per season, games started per season, and the probability of making the pro bowl. Estimation is via Tobit regression. Models 1 and 4 are censored at 0 and 1, while models 2 and 3 are censored at 0 and 16. Includes the 1991-2004 draft classes, with performance observed in the 1991-2008 seasons. N=221 targets and 3,409 other players. Standard errors in parentheses. *** p<.001, ** p<.05, * p<.01.
(1) Pr(In NFL) -0.00831*** (0.0006) 0.00000929*** (0.0000019) 0.0266 (0.055) 1.637*** (0.043) 0.649*** (0.013) 3630 0.1656 (2) Games/Season -0.0601*** (0.0046) 0.0000409*** (0.000016) 0.12 (0.42) 14.03*** (0.29) 5.864*** (0.081) 3630 0.0597 (3) Starts/Season -0.0959*** (0.0042) 0.000166*** (0.000015) 0.375 (0.38) 11.02*** (0.26) 5.177*** (0.08) 3630 0.0928 (4) Pr(Pro Bowl) -0.00938*** (0.001) 0.0000192*** (0.0000037) 0.0157 (0.07) -0.255*** (0.051) 0.613*** (0.034) 3630 0.1838

Coeff Draft Pick Draft Pick Squared Traded-for Dummy Constant /Sigma Observations R-squared

Figure S.1 Chart Convergence Excludes trades involving players or future picks. Deviation is the mean absolute error from The Chart, as estimated by the Weibull function above. Error is measured as the difference between the top pick acquired by the team trading up and the expected top pick acquired, according to The Chart, based on the rest of the picks involved in the trade. Units are in Weibull terms, with the top pick designated at 1.0. To ease interpretation: .02 equates to the 111th pick in the draft. The price deviation trend is a quadratic fit.

Price Deviation

Price Dev. Trend

Volume

Mean Abs Dev from 'Chart'

.04

.06

.08

1983

1988

1993

1998

2003

2008

Year

10

Trades per Year

.02

20

30

Figure S.2 Draftee Compensation, by Draft Order Average salary cap charge during a players first four years, by draft order. Includes the 19942008 draft classes (all which have been observed for four years since the advent of the salary cap). Includes years in which a player is not in the league. Reported in 2008 dollars.
10

Fitted values (lowess)

0 1

Average Cap Charge, millions 2 3 4 5 6 7 8

33

65 Draft Pick

97

129

161

Figure S.3 Memory Function in Performance-Value Model The weight given to performance in a players previous five years. Estimated from compensation in the players sixth through eighth year in the league. Weight is one in the most recent year, by construction.

Value relative to most recent year

.25

.5

.75

year8 year7 year6 t-5 t-4 t-3 t-2 t-1

Player Season

10

Figure S.4 Mean and Standard Deviation of Surplus Value, by Draft Pick Sample includes drafted players in their first five years, 1994-2008. Observations are the mean and standard deviation of the surplus value, at the player-season level, for players drafted at each pick. Reported in 2008 dollars.

Mean
4

Standard Deviation

Performance Value

0 1

33

65

97

129

161 1

33

65

97

129

161

Draft Pick

11

Figure S.5 Convexity of Performance Value The distribution of performance value in the Draftee sample. Performance values are estimated from the free-agent market in player years six through eight, based on the players performance in the previous five years. This is the distribution of those values after they are applied to players in the draftee sample, which consists of drafted players in their first five years.
12

Performance Value (millions)

0 0

10

10

20

30

40

50

60

70

80

90

100

Percentile

12

Figure S.6 Spline Regressions and Surplus Quantiles Regressing estimated surplus value (performance value compensation cost) on a linear spline of the draft order. The spline is knotted between rounds, so that the x variables reflect the estimated slope, i.e., .the change in surplus value, during a draft round. Splines are estimated for the mean, 25th, 50th, and 75th percentiles. Sample is limited to all drafted players, excluding punters and kickers. Reported in 2008 dollars.

Surplus, millions

-2 1

-1

33

65

97

129

161

193

225

Draft Pick
Mean p75 p50 p25

13

Figure S.7 Surplus Value Calculated Using Only Offensive Lineman Summary lowess curves for surplus value (performance value less compensation) in the players first 5 years. This is a summary for offensive lineman only, based on a performance model that was estimated on offensive lineman only. Underlying observations are player-seasons, 19942008. Reported in 2008 dollars.

Offensive Lineman Only


Surplus Value (millions)
.9 .5 1 .6 .7 .8

33

65

97

129

161

193

225

257

Draft Pick
Stage-1 compensation model was also estimated only for OL. Lowess curve, bw(.8)

14

Figure S.8 Surplus Value Calculated Using Only Wide Receivers Expected surplus, using only wide receivers. As above, surplus is the difference between performance value and compensation cost. Performance is measured by yards receiving.
800000

Expected Surplus

0 1

200000

400000

600000

33

65

97

129

161

Draft Pick

15

Figure S.9 Portfolio Analysis: Gains From Trading Down The net impact of all hypothetically possible 2-for-1 trades in the 1991-2004 drafts. This analysis focuses on first-round picks, i.e., each first-round pick and the trades down that are available from that position. Trades are determined by The Chart we consider all possible 2-pick combinations whose total chart value is 90-100% of the value of the first-round pick. Impact is assessed using two measures of performance games started and pro bowls, both measured over the players first five seasons. For every draft-pick position and year, there is an average of 28.5 portfolios assessed, each representing a potential 2-for-1 trade from that position. Means and standard errors are estimated separately for each draft-pick position, clustering on draft-year. Positive values indicate gains from trading down.

Starts per Season


1.5 9

Pro Bowls per Season

8 16 24 32 Original Pick (Trading Down From Here) Mean +/- 1SE

-.5 1

-3

.5

8 16 24 32 Original Pick (Trading Down From Here) Mean +/- 1SE

16

Figure S.10 Portfolio Analysis: Gains From Trading Down (Compensation-constant) The net impact of all hypothetically possible 2-for-1 trades in the 1991-2004 drafts. This analysis focuses on first-round picks, i.e., each first-round pick and the trades down that are available from that position. Trades are determined by compensation we consider all possible 2-pick combinations whose total rookie compensation is 90-100% of the compensation of the firstround pick. Impact is assessed using two measures of performance games started and pro bowls, both measured over the players first five seasons. For every draft-pick position and year, there is an average of 28.5 portfolios assessed, each representing a potential 2-for-1 trade from that position. Means and standard errors are estimated separately for each draft-pick position, clustering on draft-year. Positive values indicate gains from trading down.

Starts per Season


1.5 9

Pro Bowls per Season

8 16 24 32 Original Pick (Trading Down From Here) Mean +/- 1SE

-.5 1

-3

.5

8 16 24 32 Original Pick (Trading Down From Here) Mean +/- 1SE

17