Untalented but Successful Olivier GERGAUD Univ. de Reims – Univ. Paris 1 Joint with Vincenzo VERARDI CRED, Namur
Réactions : « On aimerait que Libé ouvre ses colonnes à des économistes qui bossent sur de vrais sujets, plutôt que de s'amuser dans la cour de récré », Victoria Beckham « Laisser les économistes jouer dans la cour d'école c'est aussi leur permettre de repenser la théorie classique […] Il n'y a qu'en se posant des questions iconoclastes que la science économique avance », Captain Cavern
Theory of Superstars: Fundamentals • Rosen (1981) : talent based Small differences in talent can lead to large earnings differences and eventually to the emergence of superstars. Inferior talent is not a substitute for superior talent. • Adler (1985) : talent independent Stars can even emerge among equally-talented individuals. Network effects.
Theory of Superstars: Fundamentals Rosen (1981) : talent based Inferior talent is not a substitute for superior talent. Mother
Theory of Superstars: Fundamentals Adler (1985): talent independent (network) Mother
Superstars phenomenon :Adler (2005) Adler (2005) p.3, “Superstars [...] are individuals who attain considerable prominence and success in their field and whose earnings as a result are significantly greater than the earnings of their competitors”.
Superstars phenomenon :Adler (2005) "Superstars attain earnings significantly greater than their competitors" Meaning: For a similar level of talent, some individuals will be valued more than others.
Talent: definition Talent \Tal"ent\, n. [F., fr. L. talentum] 1. Among the ancient Greeks, a weight and a denomination of money equal to 60 min[ae] or 6,000 drachm[ae]. Webster's Revised Unabridged Dictionary (1913)
Talent: definition (cont.) 4. Intellectual ability, natural or acquired; mental endowment or capacity; skill in accomplishing; a special gift, particularly in business, art, or the like; faculty. Webster's Revised Unabridged Dictionary (1913)
Motivation • Are Rosen-MacDonald and Adler Theories of Superstars complementary?
Empirical Evidence : Fuzzy • Emergence of superstars may be… • … The reward for a greater talent: à la Rosen (1981) • Hamlen (1991) • Hamlen (1994) • Lucifora and Simmons (2003) • … Independent of talent: à la Adler (1985) • Blass (1992) • Chung and Cox (1994)
(Music) Link quality of voice and sales but find no evidence of Rosen’s effect Empirical Evidence : Fuzzy • Emergence of superstars may be… • … The reward for a greater talent: à la Rosen (1981) • Hamlen (1991) • Hamlen (1994) • … Independent of talent: à la Adler • Blass (1992) • Chung and Cox (1994)
Empirical Evidence : Fuzzy • Emergence of superstars may be… • … The reward for a greater talent: à la Rosen (1981) • Lucifora and Simmons (2003) : (Soccer) convex structure of rewards but measure of talent questionable, endogenous (# of goals) : weak evidence of a Rosen effect “If we think of the top three players as the best players in Italian football measured by our goals per season indicator, then […] our results are consistent with Rosen’s (1981) theory of superstars”, Lucifora and Simmons (03)
Empirical Evidence : Fuzzy • Emergence of superstars may be… • … The reward for a greater talent: à la Rosen (1981) • Hamlen (1991); • Hamlen (1994). • … Independent of talent: à la Adler (1985) • Blass (1992) (Baseball) finds evidence against Rosen: experience is more important than talent
Empirical Evidence : Fuzzy • Emergence of superstars may be… • … The reward for a greater talent: à la Rosen (1981) • Hamlen (1991); • Hamlen (1994). • … Independent of talent: à la Adler (1985) • Chung and Cox (1994) (Music) find evidence against Rosen, there is a snowball effect on CD sold, success is the result of a prob. mechanismwhich predicts that “artistic outputs will be concentrated among a few lucky individuals”
How to test theories of superstars? Use data where: • Talent is objectively measurable • There is no unobserved heterogeneity • Superstars exist • Rarity can be separated from talent • Earning is measured with precision • Role of agents or managers is negligible In other words, use quasi-experimental data
Solution: Quasi-experimental dataset • Pokemon TCG
(i.e. ) Solution: Quasi-experimental dataset • Pokemon TCG • All characteristics of monsters are available • Talent is objectively summarized in a one-dimensional indicator • Superstars are present • Trading prices are available • Objective rarity is provided • There are no agents …
Supply of the cards • First-hand market: 19 decks in 2000 • Not included: 65% • [1,4] times: 32% • > 4 times: 3% • No individual prices. • Bad cards >>> Very good cards. • Second-hand market: • Retailers, specialized games shops or websites • Individual cards Individual prices. • SCRYE: Median price charged by retail outlets across the US and Canada.
Attack Damage 12 A sample card Name Hit Points Type Power Pokemon Data Attack Cost Weakness and Resistance Level Number
76 A sample card Name Hit Points Type Power Pokemon Data Attack Cost Attack Damage Weakness and Resistance Level Number
Advantages of our dataset • Talent (Level): • fully observable, • totally objective, • explicitly provided in the cards. • Supply : • exogenously controlled by a single firm. • Price : • measure of success, • proxy for consumers' preferences.
20$ 13$ 41 52 Rosen effect Convexity: « Greater magnification of the earnings-talent gradient increasing sharply near the top of the scale. »
1.5$ 0.5$ 12 20 Adler effect? Snowball effect: «Superstars may emerge even among equally-talented individuals»
The empirical procedure • First step: • Estimate a Robust Hedonic Price Function for the Pokémon TCG • Second step: • Detect outliers via a simple (robust) analysis of the residuals and determine the fair price for all individuals • Third step: • Graph the overprized individuals • Fourth step: • Estimate a Reweighted Hedonic Price Function for the Pokémon TCG
Superstar Superstar P Superstar Classical regression Masking effect Swamping effect OLS Z
Superstar Superstar P Superstar Robust regression Robust regression Z
LTS regression : Intuition • Decide a percentage of trimming (α%) that indicates the degree of resistance to outliers. • Run OLS on all samples having (1-α%) of the data • Choose the regression with the best fit
LTS regression: example • N = 100 & trimming 25% • Number of sub samples: • Number of sub samples: 2.42 1023
2.42 1023?? • Fast-LTS algorithm: • based on the min # of • sub samples guaranteeing • at least one non-corrupt sample
The Specification Convex Observed price Creature's characteristics Card's setting Supply conditions Rarity Rosen effect
The Specification Estimate robust residuals (ri ) by LTS
Overpricing can be estimated by: The Specification Estimate robust residuals (ri ) by LTS
How to test Rosen effect? • Run a reweighted least squares hedonic price function • Check if we observe a convexity
Results … …
Predicted by Rosen Predicted by Adler Charizard Blastoise Graphically Price Venusaur Gyarados Talent
Identifying Superstars (year 2000) • Who is sold at more than 25% of its fair price?
Conclusion: Innovations • Original Dataset: Collectible cards • Robust estimation methods • Original Results…
Conclusion: Original Results • Rosen (81) and Adler (85): complementary • Untalented but Successful: why not !!! • “We are all made of stars”
Untalented but Successful Olivier GERGAUD Université de Reims Joint with Vincenzo VERARDI CRED - Namur Talented - Unsuccessful Electrode On the job market ! Untalented - Successful Pikachu