Naïve Herding in Rich-Information Settings

Beginning with Abhijit V. Banerjee (1992) and Sushil Bikhchandani, David Hirshleifer, and Ivo Welch (1992), a theoretical literature has explored rational inference in social-learning settings. In the simplest model, a sequence of people each in turn choose one of two options, A or B, with each person observing all of her predecessors’ choices. They have common preferences over the two choices, but do not know which is better. Rather, they receive independent and equally strong private binary signals about the right choice. In this setting, rational agents herd. Once the pattern of signals leads to two more choices of one action than the other, all subsequent people ignore their signals and take that same action. This happens because two A choices (say) on the trail of equal numbers of A and B choices reveal (given the convenient simplification that people follow their own signal when indifferent) two signals favouring A; each subsequent mover, even with a B signal, thinks A a better bet. Generalizing this result, the rational-social-learning literature finds that when action and signal spaces are both finite, and each signal is imperfect, rational people eventually “herd” on an action because after a while an “ information cascade”