Big data in macroeconomic forecasting: On the usefulness of knowledge discovery in databases

Traditional methods in econometrics have problems in analyzing efficiently "big data" or large combinatorial problems respectively. The forecasting profession in macroeconomics frequently deals with problems of that kind in applying various forms of factor analysis. Whereas, as recent literature shows, factor analysis reveals some problems in handling larger data sets. In this paper it is argued how useful the utilization of techniques and methods from the field of knowledge discovery in databases (KDD) can be. Whereas the term KDD emphasizes the aspect of extracting useful information out of "big data" or the solving of large combinatorial problems respectively. In this paper the usefulness of KDD is illustrated in by applying a genetic algorithm and artificial neural networks for the purpose of forecasting macroeconomic aggregates such as industrial production and inflation. Both methods can be categorized as methods from the field of artificial intelligence and are combined in a way that a GA serves as a tool for model selection and an ANN serves for the generation (estimation) of forecasts. It is shown that the utilization of KDD can be effective in forecasting macroeconomic time series.