Working Paper: CEPR ID: DP13483
Authors: Emre Ozdenoren; Heidrun C. Hoppewewetzer; Georgios Katsenos
Abstract: This paper offers a model of experimentation and learning with uncertain outcomes as suggested by Arrow (1969). Investigating a two-player stopping game, we show that competition leads to less experimentation, which extends existing results for preemption games to the context of experimentation with uncertain outcomes. Furthermore, we inquire about the extent of experimentation under two information settings: when the researchers share information about the outcomes of their experiments and when they do not share such information. We discover that the sharing of information can generate more experimentation and higher value for a relatively wide range of parameters. We trace this finding to the stronger ability to coordinate on the information obtained through experimentation when it is shared. Our model allows to shed light on recent criticism of the current scientific system.
Keywords: stopping game; experimentation; learning; preemption; multi-armed bandit problem
JEL Codes: D83; O31
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| competition (L13) | less experimentation (C90) |
| information sharing (O36) | more experimentation (C90) |
| more experimentation (C90) | higher welfare (I31) |
| not sharing information (D82) | increased uncertainty about potential outcomes (D81) |
| increased uncertainty about potential outcomes (D81) | inhibit experimentation (C90) |
| common learning (A39) | more experimentation (C90) |
| private learning (H42) | less experimentation (C90) |