Do Individuals Actually Recognize Cascade Behaviors of Predecessors?
In class we have learned about information cascades and the properties upon which they operate. One of the implicit assumptions we make in this model is that individuals partaking are aware of the cascade behavior of others. This got me thinking – is this a realistic assumption to make? After all, there is a lot of ignorance in the world and there are a lot of people that solely focus on their own actions. It turns out that in November of 2006 T. Grebe, J. Schmid, and A. Stiehler published a study entitled “Do Individuals Recognize Cascade Behavior of Others? An Experimental Study. In this study they examine whether or not “agents” perceive the cascade behavior of their predecessors by examining the agents’ individual price patterns. The price patterns are representative of the maximum prices the agents are willing to pay to participate in a prediction game. Ultimately the authors find that more than two thirds of the participants do not recognize the cascade behavior of their predecessors, an interesting find that seems to go against the information cascade model we studied in class.
The aforementioned predecessors in this study were “artificial agents” who followed a simple counting rule and by definition never made an error in their maximum prices (bids) to participate in the prediction game. The authors then ask the subjects of the study to state their predictions and maximum prices for all possible decision situations. The result is the observation of complete individual price setting patterns. By excluding the error-making behavior of predecessors, the authors are able to address the question of whether or not individual recognize cascade behavior of others in isolation.
What the authors found was that most subjects predict according to theory (and to simple counting) but many submit increasing maximum prices the more coinciding predictions of predecessors they observe, regardless of whether additional information is revealed by these predictions. It is important to note that since the artificial agents in this study can never make an error, beliefs about the predecessors that are created are different from those in experiments with human players. Due to this, the behavior observed in this experiment differs in some aspects from behavior reported in other cascade studies.
The authors posit an alternative behavioral hypothesis to the information cascade hypothesis we learned in class. We learned in class that individuals update information according to Bayes’ rule and take cascade behavior of others into account. The behavioral hypothesis states that individuals update information according to Bayes’ rule but do not recognize cascade behavior of others.
Ultimately the authors concluded that their findings support the alternative (behavioral) hypothesis and that individuals do not recognize the cascade behavior of others. Only 18% of the participants set their prices in accordance with the information cascade model (commonly referred to as the BHW model in the article). The striking difference between the experimental environment in this study and reality is that the predecessors made no errors. This is of course wildly unrealistic as no one is right 100% of the time. The authors state however that if the individuals in the study were unable to recognize cascade behavior in a simple setting where the predecessors were always right, then it is unlikely that they will do so in a real-life environment with all the complexities that will be added to the system when the predecessors are humans.
The findings of this study were particularly striking in that they seemed to go against one of the underlying assumptions of the information cascade model that we have learned about in class: individuals are able to recognize cascade behavior of those who made their decisions before them. However, the predecessors in this study were unable to make errors, and there is no real way to say that the subjects in the study would act the same and fail to recognize cascade behavior if they knew that their predecessors were also human and thus inherently subject to error. Furthermore, there is a distinction that is made in the article that acting based upon the predecessors’ actions and being cognizant of the fact that these predecessors are acting based upon cascade behavior are two different things. It will remain to be seen with further research if this study sets the groundwork for calling into question the validity of the BHW (information cascade) model.
Source: http://sfb649.wiwi.hu-berlin.de/papers/pdf/SFB649DP2006-079.pdf