This book describes a series of laboratory experiments (witha total of 167 independent subjects) on forecastingbehavior. In all experiments, the time series to beforecasted was generated by an abstract econometric modelinvolving two or three artificial exogenous variables. Thisdesignprovides an optimal background for rationalexpectations and least-squares learning. As expected, thesehypotheses do not explain observed forecasting behaviorsatisfactorily. Some phenomena related to this lack ofrationality are studied: Concentration on changes ratherthan levels,underestimation of changes and overvaluation ofvolatile exogenous variables. Some learning behavior isobserved. Finally, some aspects of individual forecasts suchas prominence of "round" number, dispersion, etc.,arestudied.