The canonical model of human behavior in economics, which rose to prominence in the last half of the twentieth century, is perfect rationality. This model is appealing because it makes strong predictions based on parsimonious assumptions. The standard approach in contemporary economics is to modify perfect rationality, e.g. by assuming that agents have different information or act under institutional constraints. A more drastic modification is to assume bounded rationality. However, realistic models of bounded rationality tend to be complicated, ad hoc, and intractable.
There is, however, an alternative approach that is equally parsimonious, and in some cases also makes strong predictions. This approach is to assume that rationality is so bounded as to be almost nonexistent. I will illustrate this approach with a model for price formation in which agents use the current price as a reference point, but otherwise behave randomly. Surprisingly, when tested against data from the London Stock Exchange, the model performs well, yielding predictions for price variability and market impact that rival the best predictions anywhere in economics. We believe the reason this model works is because it is possible to accurately model the market institution used for price formation, incorporating the feedback between agent behavior and price setting. This suggests an alternative approach to economics: Start with random agents and then add a little intelligence.