Forecasting Inflation Under Varying Frequencies

This paper seeks to determine the impact of monthly and annual data frequencies on the accuracy of inflation forecasts attainable via econometric and subspace-based methods. The application considers food inflation across short and long run horizons in Colombia, a country with an inflation targeting regime. The data includes all 54 components of the food consumer price index (CPI) in Colombia from Jan. 1999 – Oct. 2012, and the study forecasts the food CPI, and inflation using the parametric and nonparametric techniques of ARIMA, Exponential Smoothing (ETS), Holt-Winters (HW) and Singular Spectrum Analysis (SSA). We find that when forecasting the index, ARIMA forecasts are on average best, whilst for monthly inflation forecasting SSA is comparatively better and for annual, the results vary between SSA and ARIMA. These statistically significant findings give policy makers an option to select an apt forecasting model which suits their requirements.