Boosting as a Forecasting Method Using Large Data Sets: Evidence for the USA, the Euro Area, and Germany

• Activity Number: 248
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Teresa Buchen*+ and Klaus Wohlrabe
• Companies: Ifo Institute for Economic Research and Ifo institute for Economic Research
• Keywords: boosting ; forecasting ; large datasets

Abstract

Recently, there has been rising interest in the use of large datasets for forecasting since both the availability of data and the computational power to handle them have strongly increased. Since traditional econometric methods are not suitable for incorporating a large number of predictors, new methods were developed. Boosting is a prediction method for high-dimensional data that was invented in the machine learning community. It is a stagewise additive modelling procedure, which, in a linear specication, becomes a variable selection device that sequentially adds the predictors with the largest contribution to the fit. The few related studies that exist in the macroeconometric literature are all restricted to U.S. data. By analysing large datasets for the USA, the euro area and Germany, this study evaluates the forecasting accuracy of boosting with respect to several key economic variables. We find that boosting mostly outperforms the benchmark, but the gains are largest in the medium run. Moreover, the forecast errors decrease when selecting a less noisy subset of the data. So the quality of data greatly determines the forecasting performance of boosting.