Application of Machine Learning Methods for Forecasting the Consumer Price Index

Abstract

The article analyzes the effectiveness of machine learning methods for forecasting the consumer price index, which is the most important macroeconomic indicator of inflation and a key statistical indicator for governments and central banks in the process of assessing price stability. The research uses daily data on a large number of goods and services (more than 12 million units) from June 2020 to May 2022 as an information base. The difficulty of using data for forecasting lies in the presence of gaps in observations. Using special methods for processing gaps, resampling and eliminating outliers, data on individual goods are converted to continuous time series with equally spaced observations. The following machine learning methods are used for forecasting: PyAF, StatsForecastAutoARIMA, Prophet, LSTM recurrent neural networks. The article also uses machine learning methods to analyze the impact of such factors on inflation as the price of Brent crude oil, the rate of the Central Bank of the Russian Federation, the dollar to ruble exchange rate, the level of inflation for the month according to Rosstat data, GDP, the balance of trade, and examines the possibility of using these factors in forecasting models.