Robust thresholding for Diffusion Index forecast

In this paper we propose a new methodology in improving the Diffusion Index forecasting model (Stock and Watson, 2002a, 2002b) using hard thresholding with robust KVB statistic for regression hypothesis tests (Kiefer et al., 2000). The new method yields promising results in the context of long forecasting horizons and existence of serial correlation. Numerical comparison indicates that the proposed methodology can improve upon the existing hard thresholding methods and outperform the soft thresholding methods (Bai and Ng, 2008) when applied to a real data set that forecasts eight macroeconomic variables in the United States.