Hffi TB^v / of DEWEY M.I.T. LIBRARIES - Digitized by the Internet Archive in 2011 with funding from Boston Library Consortium Member Libraries http://www.archive.org/details/pathologicalmodeOOsmit » HB31 Dewey .M415 AO , working paper department of economics PATHOLOGICAL MODELS OF OBSERVATIONAL LEARNING Lones Smith Peter S^rensen 94-24 July 1994 massachusetts institute of technology 50 memorial drive Cambridge, mass. 02139 PATHOLOGICAL MODELS OF OBSERVATIONAL LEARNING Lones Smith Peter S^rensen 94-24 July 1994 L«icF-fTSINSTITUTE OCT 2 1 1994 Pathological Models of Observational Learning" Lones Smith* and Peter Sorensen* Department of Economics Massachusetts Institute of Technology MA Cambridge, 02139 (Preliminary version. Comments encouraged.) July 27, 1994 Abstract This paper systematically analyzes the observational learning paradigm of Banerjee (1992) and Bikhchandani, Hirshleifer and Welch (1992). We first relax the informational assumption that is the linchpin of the "herding' re- sults, namely that individuals' private signals are uniformly bounded in their strength. We then investigate the model with heterogeneous preferences, and discover that a 'twin' observational pathology generically appears: Optimal learning may well lead to a situation where no one can draw any inference at all from history! We also point out that counterintuitively. even with a constant background "noise" induced by trembling or crazy individuals, public beliefs generically converge to the true state of the world. All results are cast within a simple dynamic mathematical framework that is (i) rich enough to describe a rich array of observational learning dynamics; and (ii) amenable to economic modifications that hinder or abet informational transmission, and sometimes permit full belief convergence to occur. "The authors wish to thank participants at the MIT theory lunch, in particular Abhijit Banerjee. for their comments on a somewhat related paper. We are especially grateful to Chris Avery, who was very instrumental at an early stage in this project. All errors remain our responsibility. Sorensen gratefully acknowledges financial support for this work from the Danish Social Sciences Research Council. ^e-mail address: lonesClone3.n1it.edu 'e-mail address: petersorfimit edu .