Comparative Analysis of Regional Inflation Forecasting Models

The purpose of this study is to compare approaches to forecasting the monthly level of CPI y/y in the regions of the Volga Federal District using time series models and machine learning methods. Despite the fact that this topic is among the traditional studies – there are many theoretical and empirical works devoted to the research under consideration – it always arouses interest. This study attempts to compare both approaches and select the most appropriate and efficient models for predicting the regional general price level index. The work also contains the use of a combined approach, which is based on the combination of both methods. The results show that machine learning models provide more stable and accurate forecasts than econometric models – especially over long forecasting periods (6 months or more), according to the Root Mean Squared Error (RMSE) values. However, for a number of regions, we found evidence of the effectiveness of time series models on the short term – for several regions, different specifications of extended autoregressive models perform better than the machine learning model approach when forecasting for 1 and 3 months. In addition, the simple naive model shows more accurate results when forecasting the monthly level of CPI y/y in the Republic of Mordovia compared to both approaches. The results of the combined approach are comparable to the forecasts of machine learning models and more often provide more accurate forecasts for 12 and 24 months.