Data Mining for Big Data Macroeconomic Forecasting: A Complementary Approach to Factor Models

4. Conclusions

Against the background that methods for efficient use of big data sets become increasingly important in applied macroeconomic forecasting literature we presented a forecast model selection approach based on a GA which tries to overcome problems of alternative quantitative methods, e.g., factor analysis and artificial neural networks. The need for new methods is caused by using big data sets for which the use of GAs (as a typical data mining method) seems to be appropriate. Starting from a big data set with typical macroeconomic variables such as German leading indicators and key indicators our goal was to make forecasts for the German industrial production, a long maturity bond, inflation and unemployment. We employed a GA to optimize forecast models. Our results meet all forecasting requirements and stress the advantages of our approach as opposed to alternative methods.