This book describes a series of laboratory experiments (with a total of 167 independent subjects) on forecasting behavior. In all experiments, the time series to be forecasted was generated by an abstract econometric model involving two or three artificial exogenous variables. This designprovides an optimal background for rational expectations and least-squares learning. As expected, these hypotheses do not explain observed forecasting behavior satisfactorily. Some phenomena related to this lack of rationality are studied: Concentration on changes rather than levels, underestimation of changes and overvaluation of volatile exogenous variables. Some learning behavior is observed. Finally, some aspects of individual forecasts such as prominence of "round" number, dispersion, etc., are studied.