Performance of Partial Least Squares models in Forecasts of Inflation

This paper aims to compare the forecast performance of Partial Least Squares (PLS) models with Auto-Regressive and Principal Components (PC) models. Ordinary Least Squares (OLS) and Ridge estimations are used to approximate the coefficients. The main evaluation is done by applying Relative Mean Squared forecast Errors. It is shown that the Static PLS model often outperforms the other approaches and the Dynamic PLS model shows an improving results with an increase in the amount of steps ahead. It is also noted that the Ridge estimation enhances the prediction power of the PC approach compared to the OLS estimation.