Investigating the effect of changes in Signals, Noise and Agent Types in Cascade Models

Smith and SA¸renson (2002) depart from Bikhchandani, Hirshleifer and Welsh’s (1999) herding framework in three ways; non-discrete signals, noise and multiple rational agent types. They refer to the standard model of a single type with no noise as a “herding†model, and differentiate themselves by referring to the more general model as a “cascade.†They find that differing preferences lead to “confounded learning†whereby an incorrect equilibrium can arise regardless of the accuracy of the signal space, due to alternate actions taken due to preferences. In this paper we break down the three alterations from the BHW (1999) model, and analyze through agent based techniques the effect that each one has on the outcome, in an effort to determine which of the three different specifications has the greatest impact on herding and learning in the cascade model.