A comprehensive look at the forecast performance with robust forecast comparison

Abstract

Forecasting economic and financial variable is important that success or failure of monetary policy or hedging strategies depends heavily on the accuracy of the forecasts. Hence, decision on the model to be used in forecasting is crucial. However, until now, the forecast superiority of one model relative to others has been dependent on the loss function that was specified. To circumvent this problem, this paper uses the concepts of general-loss forecast superiority and convex-loss forecast superiority tests introduced by Jin et al.(2016). They developed the forecast evaluation tests based upon a mapping between stochastic dominance and forecast superiority. Also, this paper shows that the combination of the LASSO-based approaches and dynamic factor model delivers superior forecasts than a variety of competing models suggested by many studies. There have been a great deal of evidence that combining models can reduce the forecast error using moment-based criteria. As an extension of those results, we investigate the performance of forecasting combination based on the entire distribution of forecast errors.