Econometric and Statistical Data Mining , Prediction and Policy - Making

How to formulate models that work well in explanation, prediction and policy-making is a central problem in all fields of science. In this presentation, I shall explain the strategy, our Structural Econometric Modeling, Times Series Analysis (SEMTSA) approach that my colleagues and I have employed in our efforts to produce a macroeconomic model that works well in point and turning point forecasting, explanation and policy-making. Data relating to 18 industrialized countries over the years, taken from the IMF-IFS data base have been employed in estimation and forecasting tests of our models using fixed and time varying parameter models, Bayesian posterior odds, model combining or averaging, shrinkage, and Bayesian method of moments procedures. Building on this past work, in recent research economic theory and data for 11 sectors of the U.S. economy have been employed to produce models for each sector. The use of sector data and models to forecast individual sectors’ output growth rates and from them growth rates of total U.S. output will be compared to use of aggregate data and models to forecast growth rates of total U.S. output. As will be seen, IT PAYS TO DISAGGREGATE in this instance. Last, a description of some steps underway to improve and complete our Marshallian Macroeconomic Model of an economy will be described.