Recent advances in computing technology and data availability have sharply reduced the cost of real-time forecasting of the economy using large numbers of time series data. At the same time, there is considerable evidence that economic forecasting systems have had significant episodes of instability, a recent example being the poor performance of most models entering the 1990 to 1991 recession and the subsequent slow recovery. The purpose of this research is to develop and analyze forecasting methodologies when predictive relationships are changing over time. Several large data and rich data sets will be used to assess the performance of various adaptive forecasting systems. These include data from the postwar U.S., data from the U.S. before World War II, and data from Canada and Israel. This research is important because it holds the promise of improving significantly economic forecasting. An important application would be in the area of improving long-term energy forecasting and assessing global environmental change.
