Direct vs. iterated forecasts for the U.S. GDP growth rate.

Abstract

The aim of this study is twofold. First, we investigate the predictive role of some economic and financial variables for the U.S. economic activity at short-term forecast horizons. We do this by generating recursively out-of-sample forecasts in the context of simple autoregressive models. Second, the main goal is to compare two forecasting methods, namely the iterated and the direct approach, regarding their forecasting ability for the U.S. GDP growth rate. We initially review the voluminous literature on this topic. We then set the theoretical framework for our empirical analysis. Afterwards, we undertake an empirical analysis using quarterly data for a span of up to 41 years (1973-2013). Empirical results indicate that none of the candidate variables is systematically an efficient predictor of the U.S output growth. Finally, we find that direct forecasts are superior to iterated forecasts at all forecast horizons.