Robust Forecasting with Scaled Independent Component Analysis

In this paper, a scaled independent component analysis (sICA) method is proposed for finding potential factors with more predictive power. The core idea is to improve the predictive effect of the model by giving more weight to those variables with stronger predictive power before estimating the independent components. Finally, we apply our method to study the data from the Federal Reserve Monthly Database for Economic Research (FRED-MD) which is a large macroeconomic database. The results of the data analysis show that, in general, the sICA method has better forecasting performance