The Impact of Disaggregation on the Forecasting Accuracy of Inflation

Annotation This master’s thesis examines how data disaggregation affects forecasting accuracy of Armenian CPI inflation. With different methods and models, we try to understand how disaggregation of Armenian Inflation can improve forecasts over the short-term horizons. We want to understand whether it is better to forecast CPI inflation directly or instead combine the forecasts from the disaggregated models or include the components to forecast the aggregated CPI inflation. Can information contained in the components help to forecast CPI? To understand the above-mentioned idea, we estimate different models with the aggregated and disaggregated CPI inflation. At first, we compare the forecasting results from the aggregated and disaggregated models. Then we include disaggregated variables to forecast the aggregate. We also use the information contained in the subcomponents of the CPI (factors estimated from subcomponents) to forecast the aggregated CPI inflation. Besides these, we examine how the level of disaggregation influences on the forecasting accuracy of the models. For the CPI and its components, we apply ARMA and unrestricted VAR models. In the models, where macroeconomic variables are included, we apply BVAR and factor-augmented VAR (FAVAR is also applied when we include disaggregated information in the aggregate model). The goal of our work is to understand how we can use the components or the information contained in the components to efficiently estimate and forecast the aggregated CPI inflation. Is it better to forecast CPI directly, or instead use the disaggregate variables or disaggregate information to forecast the aggregate?