The analysis of nonstationary time series, unit roots and cointegration has developed dramatically over the last 12 years. The papers here variously describe new methods, they evaluate methods, they provide useful overviews and they show detailed implementations that are helpful to practitioners. However these papers show not only developments in methods of estimating long-ran relations (for seasonally unadjusted as well as adjusted data, for instance), but also how the whole subject matter is broadening out to have a profound affect on econometric analysis in general. Michael Clements and David Hendry's discussion of economic forecasting is based around an integral understanding of integration and cointegration including an analysis of whether one should use differenced models. Fabio Canova, Mary Finn and Adrian Pagan's analysis of evaluating real business cycle models and Steven Durlauf and Mark Hooker's analysis of Cagan's hyperinflation model equally work from an understanding of the different time series properties to be found in nonstationary versus stationary data. The concepts of integration and cointegration are here to stay although some of the demands made on the data are very high as we shall discuss later.
The first paper by Michael Clements and David Hendry is a substantive and innovative contribution to analysing economic forecasting, looking at the various sources of error such as 'parameter non-constancy, estimation uncertainty, variable uncertainty, innovation uncertainty, and model misspecification', and trying to organise these into a coherent theory of economic forecasting. This approach appears to have great potential and certainly leads to a better appreciation of the use of intercept adjustments to improve forecasting. One could see this sort of approach being used also for a well-organised analysis of the many different types