Multi-horizon comparison of multivariate inflation forecasting

This paper applies the multi-horizon comparison methodology from Quaedvlieg (2019) to assess the forecasting performance of direct and iterative multivariate inflation forecasts, with both highand low lagorders. We use variousmacroeconomic indicators ina GETS restricted estimation to forecastUSinflation and show that high orderVARs on average prefer iterative forecasts, while low order VARs on average prefer the direct forecasts. Finally, we provide evidence that the best high order multivariate forecasts outperform the best low order multivariate forecastson every individual horizon(uniform superior predictive abilities). Thisimpliesthat in this setting, inflation forecasts are most accurately forecasted with a high orderVAR using an iterativeapproach.