Selection of Macroeconomic Forecasting Models: One Size Fits All?

Abstract

The main distinction between this paper and traditional approach is the assumption that variables affect the economy through different horizons. Under this alternative hypothesis, a variable considered as an unimportant detail from a short-horizon perspective may become an essential factor in a long-horizon standpoint, this paper, therefore, suggests selecting variables specific to the horizon. My findings confirm that a model that allows the variables particular to the horizon has a lower Schwarz Bayesian Information Criterion (SBIC) value than a model that does not. My outcomes also show that the vector autoregression (VAR) model in general forecasts poorly compared with my approach. Likewise, I contribute to the literature by setting predictions equal to the sample mean as a benchmark and showing that the out-of-sample forecasts of the VAR model with lag length higher than one fail to outperform the sample mean. Additionally, I select principal components derived from 190 different time series to forecast a time series as the time horizon varies. Again, the results show that some of the principal components may be more important at some horizons than at others, thus I suggest selecting the principal components in a factor-augmented VAR (FAVAR) model specific to the horizon. According to above results, I conclude that long-horizon and deep-rooted economic problems cannot be fixed with short-horizon and surface-level interventions. I also reach my argument via simulation.