A re-evaluation of the term spread as a leading indicator

摘要: Abstract Forecasting the evolution path of macroeconomic variables has always been keen interest to policy makers and market participants. A common tool used in relevant forecasting literature is term spread Treasury bond yields. In this paper, we decompose into an expectation a premium component evaluate informational content each GDP growth rate inflation various horizons. doing so, employ alternative decomposition procedures introduce Support Vector Regression (SVR) methodology from field Machine Learning, coupled with linear non-linear kernels as novel method field. Using rolling windows producing point conditional probability distribution forecasts find that neither spread, nor its components possess ability accurately forecast output or inflation. Our findings extend existing literature, since they are focused on explicit out-of-sample evaluation contrast most empirical studies produce only in-sample forecasts. To strengthen our findings, also consider several control suggested without significant qualitative differences initial results. The main innovation approach stems use Vectors methodology, introduced for first time line research out-of-sample.