A Suite of Models for CPI Forecasting

This paper reviews the suite of models the National Bank of Ukraine uses for short-term forecasting of CPI components. I examine the forecasting accuracy of the following econometric models: univariate models, VAR, FAVAR, Bayesian VAR models, and Error Correction models. The findings suggest that for almost all components there are models that outperform benchmark AR models. However, the best performing individual model at each horizon for each component differs. Combined forecasts obtained by averaging the models' forecasts produce acceptable and robust results. Specifically, the combined forecasts are most accurate for core inflation, while they can beat the AR benchmark more frequently than other types of models when it comes to the raw food price index. This study also describes relevant data restrictions in wartime, and highlights avenues for improving the current suite of models for CPI forecasting.