This paper benefitted greatly from a faculty seminar at the University of Mississippi. Discussions with Richard Boylan, Gerard Buskes, Don Cole, Ron Harstad, David Marshall, Frank Page, Hyun Park, Paul Pecorino, Marcin Pęski, William Rogerson, Bernard Sinclair-Desgagné, Jeroen Swinkels, and Min-hung Tsay were also extremely helpful. In particular, especially perceptive questions by Bernard Sinclair-Desgagné led to the results in Section 8 and contributed significantly to Section 10. Ian Jewitt also suggested numerous major improvements, only a few of which are explicitly acknowledged below. Finally, extensive and thoughtful suggestions by the referees and by the co-editor, Larry Samuelson, have greatly streamlined and improved the paper. This work was made possible, in part, by a sabbatical from the University of Mississippi in the fall of 2005. All remaining errors are mine.
Two New Conditions Supporting the First-Order Approach to Multisignal Principal–Agent Problems
Article first published online: 15 DEC 2008
© 2009 The Econometric Society
Volume 77, Issue 1, pages 249–278, January 2009
How to Cite
Conlon, J. R. (2009), Two New Conditions Supporting the First-Order Approach to Multisignal Principal–Agent Problems. Econometrica, 77: 249–278. doi: 10.3982/ECTA6688
- Issue published online: 15 DEC 2008
- Article first published online: 15 DEC 2008
- Manuscript received September, 2006; final revision received June, 2008.
- Principal–agent model;
- moral hazard;
- first-order approach;
- multiple signals
This paper presents simple new multisignal generalizations of the two classic methods used to justify the first-order approach to moral hazard principal–agent problems, and compares these two approaches with each other. The paper first discusses limitations of previous generalizations. Then a state-space formulation is used to obtain a new multisignal generalization of the Jewitt (1988) conditions. Next, using the Mirrlees formulation, new multisignal generalizations of the convexity of the distribution function condition (CDFC) approach of Rogerson (1985) and Sinclair-Desgagné (1994) are obtained. Vector calculus methods are used to derive easy-to-check local conditions for our generalization of the CDFC. Finally, we argue that the Jewitt conditions may generalize more flexibly than the CDFC to the multisignal case. This is because, with many signals, the principal can become very well informed about the agent's action and, even in the one-signal case, the CDFC must fail when the signal becomes very accurate.