Hybrid Model and Ensemble for Inflation Forecasting: A Machine Learning Approach

An inflation index is a statistical tool that measures the change in the general price level of goods and services in an economy over a specific period, typically used to gauge the inflation rate. Understanding such an index is vital because it measures the overall price level changes in an economy over time, influencing policy decisions, investments, and the real value of money, and thus, the general economic well-being of society. This article presents a study on the forecast of the official inflation index in Brazil, IPCA, using a hybrid system with bagging. The proposed model combines a linear approach (ETS) with the bagging technique applied to residual forecasting. Thirteen relevant methods mentioned in the literature were evaluated, including widely used models such as ARIMA, ETS, MLP, SVR, RF, and various hybrid combinations. The experimental results show that the proposed hybrid model (ETS+Bagging) outperformed individual models and other hybrid systems. Furthermore, the bagging approach significantly improved the results compared to the ARIMA and ETS models, as measured by metrics such as mean squared error (MSE), root means squared error (RMSE), and mean absolute error (MAE). The results indicate that the hybrid system with bagging, especially the proposed ETS+Bagging model, has the potential to enhance inflation forecasts using IPCA data.