Working Paper: CEPR ID: DP6770
Authors: Benny Moldovanu; Aner Sela; Xianwen Shi
Abstract: We study optimal contest design in situations where the designer can reward high performance agents with positive prizes and punish low performance agents with negative prizes. We link the optimal prize structure to the curvature of distribution of abilities in the population. In particular, we identify conditions under which, even if punishment is costly, punishing the bottom is more effective than rewarding the top in eliciting effort input. If punishment is costless, we study the optimal number of punishments in the contest.
Keywords: all-pay auctions; contests; order statistics; punishments
JEL Codes: D44; D82; J31; J41
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| contest design (prizes and punishments) (D44) | participant effort (C90) |
| punishments (negative prizes) (K42) | participant effort (C90) |
| rewards (positive prizes) (M52) | participant effort (C90) |
| punishments (negative prizes) > rewards (positive prizes) under certain conditions (K42) | participant effort (C90) |
| optimal prize structure (single prize) when punishment is not feasible (D40) | participant effort (C90) |
| punish the player with the lowest performance when only punishments are feasible and ability distribution has increasing hazard rate (D80) | participant effort (C90) |
| optimal structure depends on marginal costs of rewards and punishments (D40) | participant effort (C90) |
| punishing the worst performer is optimal if ability distribution is convex (H21) | participant effort (C90) |
| optimal number of punishments decreases with convexity of ability distribution (costless punishments) (H21) | participant effort (C90) |
| punishments can help exclude low-ability participants (C92) | participant effort (C90) |
| punishments remain effective even when players opt not to participate (Z22) | expected total effort (C13) |