This chapter focuses on heterogeneous gain learning and long swings in asset prices. Many asset prices deviate from their fundamental values, yielding potential long-run predictability. Asset price swings can be both short or long in duration, and their time series shows few regular patterns when one analyzes their long-range behavior. This chapter considers an underparameterized learning model with heterogeneous gain parameters and traders using differing perspectives on history. It first provides an overview of the basic model and some benchmark simulation runs before discussing the output of the model compared to actual financial time series. It then describes a range of internal mechanisms of the agents and forecasts in use and how wealth moves across them over time. It shows that learning algorithms appear to be behaving in a predictable fashion, and that interesting dynamics come from how agent wealth selects rules over time. The chapter concludes by addressing some questions for researchers working on learning in financial markets.
A case study of the residual-based cointegration procedure