Abstract

This study investigates the advantages of forecasting inflation through its disaggregated components, using data from countries such as the United States, the United Kingdom, and Colombia. The authors compare two main approaches:

1. Forecasting each component of inflation (such as food prices, energy prices, etc.) and then combining them using the corresponding weights in the Consumer Price Index (CPI) basket relevant to each country’s CPI structure.
2. Forecasting the overall inflation rate directly.

A variety of forecasting techniques are employed, including advanced statistical methods such as dimensional reduction (Dynamic Factor Models and Principal Component Analysis) and machine learning techniques (Ridge, Lasso, and Random Forest regression), as well as traditional methods like ARIMA and TAR models. These techniques are utilized to evaluate the effectiveness of forecasting 1, 3, 6, 9, and 12 months ahead across two critical periods: before and after the COVID-19 pandemic.

The findings suggest that forecasting the individual components and then combining them works just as well as, and sometimes even better than, directly forecasting the headline and core inflation. This approach can lead to more accurate inflation predictions, which is crucial for central banks aiming to maintain price stability. It also helps identify the components that are the primary drivers of observed inflation over time. Empirical comparisons of the forecasting performance of the models after the COVID-19 pandemic show that aggregating forecasts from the disaggregated reduces the forecast error mainly in the cases of headline inflation for the United States case and for the core inflation for the United Kingdom. This result indicates that forecasts constructed with granular information outperform those built with aggregated data when turbulence is present in macroeconomic and price data.