Inflation forecast with Artificial Intelligence

Abstract

Inflation is a generalized increase in prices in an economy. Small rates of inflation are natural; however, the uncertainty related to inflation volatility brings issues in monetary policies design. In Brazil, the IPCA is adopted as an inflation target; however, the use of core inflation as a target would allow to design less rigid monetary policies. In this work, we propose the construction of wavelet core inflation, since in inflationary contexts they present a better performance in the trend analysis when compared to usual core inflation. For the forecast, artificial intelligence techniques are adopted, such as neural networks. We point out that neural network make it possible to deal with highly complex problems, which cannot always be described by analytical models. Additionally, we use confidence intervals to delimit inflation forecast probable estimates. Among the main conclusions, we emphasize that wavelet core inflation had smaller confidence intervals, in addition to presenting lesser errors in the construction of the neural network. In addition, inflation forecast generated smoothed signals, allowing to identify the trend of inflation of up to twelve months.