Nonstationary time series arising from autoregressice models with roots on or near the unit circle have been an intensive subject of econometric reseach during the last twenty years. Heikki Kauppi's doctoral dissertation deals with several topics in the area of theoretical econometrics of...

Part of conventional wisdom glaned from econometric theory and the "learning" literature is that agents ahould use all the data they have for prediction. In this paper it is shown that agents can improve their prediction by throwing away data.

We consider the effects of replacing least squares learning by stochastic gradient learning in the multivariate "Cobweb" model. Are the stability conditions altered? For this model, we show global convergence of stochastic gradient learning to the unique rational expectations equilibrium...