Experimentation and project selection: Screening and learning

Abstract

JEL classification:Firms must strike a delicate balance between the exploitation of well-known business models and the exploration of risky, untested approaches. In this paper, we study financial contracting between an investor and a firm with private information about its returns from exploration and exploitation. The investor-optimal mechanism offers contracts with different tolerance for failures to screen returns from exploitation, and with different exposure to the project's revenues to screen returns from exploration. We derive necessary and sufficient conditions for private information about returns from exploration to have zero value to the firm. When these conditions fail, private information about exploration may even decrease the firm's payoff.

Optimal Contracts for Experimentation

“…Private information about project quality is studied by Gomes et al (2015) in experimentation without moral hazard, and in a different setting by Gerardi and Maestri (2012). Another possibility would be non-common priors between the principal and the agent, which would involve quite distinct considerations.…”

Information acquisition and voting with heterogeneous experts

“…Gromb and Martimort (2004) show that this dominance of multi‐expert organization persists in sequential information gathering. More recently, there has been an effort in the contract design literature to model experimentation, a sequential information gathering process; it usually focuses on one expert (See, e.g., Bergemann and Hege, 1998, 2005; Gomes, Gottlieb, and Maestri, 2016; Halac, Kartik, and Liu, 2016; Khalil, Lawarree, and Rodivilov, 2018).…”

Can Starving Start‐ups Beat Fat Labs? A Bandit Model of Innovation with Endogenous Financing Constraint*

“…11 In Halac et al (2016), the principal-agent model involves both adverse selection on agent's ability and dynamic moral hazard. Related are also the experimentation frameworks of Drugov and Macchiavello (2014), with both adverse selection and moral hazard, Bouvard (2014), who only consider adverse selection on the quality of the agent, Gomes et al (2016), who introduce two-dimensional adverse selection, and Bobtcheff and Levy (2017), where the agent has private information on the quality of learning. Outside of a pure experimentation framework, see also Gerardi and Maestri (2012), who study dynamic incentives for information acquisition under arrival of a single success, and the agent has only one work method to produce it; moreover, the moral hazard is on the agent's actions (effort or investment).…”

“…Related are also the experimentation frameworks of Drugov and Macchiavello (), with both adverse selection and moral hazard, Bouvard (), who only consider adverse selection on the quality of the agent, Gomes et al . (), who introduce two‐dimensional adverse selection, and Bobtcheff and Levy (), where the agent has private information on the quality of learning. Outside of a pure experimentation framework, see also Gerardi and Maestri (), who study dynamic incentives for information acquisition under both moral hazard and adverse selection, and DeMarzo and Sannikov (), who study private learning under moral hazard.…”