Abstract:
|
We develop an empirical methodology to obtain forecast for total inflation from aggregating individual disaggregate forecast. To do that we estimate a model for each component of the CPI basket at the most disaggregated level. Furthermore, we evaluate the relationship between each component with other potential explanatory variables such as components of the IPP index, exchange rate, some climate indicators, and economic activity indexes. Finally, using Bayesian estimation we evaluate the forecast performance of each possible model for each item, and once we identify the best model, we aggregate the individual forecasts with its respective weight in the CPI basket. We found that for some items the best forecasting model is an ARIMA while for others the additional variables have forecasting power. The aggregated forecast performs better than the one obtained from a single model for the aggregate inflation.
|