| The social learning literature analyzes an economy where a sequence of individuals are supposed to make a decision under incomplete and asymmetric information. The typical conclusion is that, despite the asymmetry of information, eventually every individual imitates her predecessor, even though she would have chosen a different action on the basis of her own information alone. In this sense, individuals rationally 'ignore' their own information and 'follow the herd'.; A central assumption of nearly all learning models is perfect information: everyone is assumed to be informed about the entire history of actions that have already been taken. In reality, individuals have imperfect information. If each individual observes the actions of only a small number of other individuals, it is not clear that 'herd behavior' will arise. My imperfect-information model provides outcomes that are quite distinct from and in some ways more extreme than the perfect-information model.; The key economic phenomenon that imperfect information captures is a succession of 'fads' starting suddenly, expiring rather easily, each replaced by another fad. Thus, imperfect-information model provides better answers to such questions as: Why do markets move from 'boom' to 'crash' without settling down? Why does a wide range of users adopt a technology more rapidly than expected and then, suddenly, replaced by an alternative? What makes a restaurant fashionable overnight and equally unexpectedly unfashionable, while another becomes the 'in place', and so on?; The challenge of better understanding the social and economic impact of learning leads me to combine theoretical and experimental research. In the laboratory, I find that imitation is much less frequent than the theory predicts. For a better understanding of the decision mechanism of the subjects, I focus on the data at the individual level. I find that among the subjects who rationally follow their predecessor there is a considerably degree of conformity with the theory, which we fail to observe in the aggregate data. Thus, I discuss an alternative theoretical explanation for the experimental results by introducing noisy individuals in the benchmark model. |
