Forecasting Inflation in Russia Using Gradient Boosting and Neural Networks

Abstract

The aim of this paper is to estimate the efficiency of forecasting inflation in Russia using machine learning methods such as gradient boosting algorithms and neural networks. This is the first paper in which long short-term memory (LSTM) and gated recurrent unit (GRU) models are used to forecast inflation in Russia. In addition, I test modified versions of gradient boosting such as LightGBM and CatBoost. With a sample of lagged inflation values, the most accurate forecasts are obtained using convolutional neural networks (CNNs) and fully connected neural network (FCNNs), and when forecasting over a twelve-month horizon, using the LSTM model, which is associated with sequential information processing and the gating mechanism in statistical data analysis. When additional macroeconomic factors are taken into account, FCNNs and the Sklearn gradient boosting model demonstrate a predictive advantage. As per the Shapley decomposition, the most informative predictors for forecasting Russian inflation are oil and natural gas prices, inflation in the euro area and the United States, retail trade turnover dynamics, and the unemployment growth.