Disaggregated Approach to Consumer Price Index Modeling: Regional Aspect

The paper solves the problem of accuracy improvement of forecasting the consumer price index (CPI), which is a key indicator of inflationary processes. It is used for monetary policy instruments. In contrast to the traditional approach, in which the forecast is made for the aggregated indicator “All goods and services,” which does not allow taking into account the heterogeneity of the dynamics of goods and services categories, the authors implemented disaggregated CPI modeling at the regional level. Forecasting was carried out at three levels: CPI as a whole, by basic components (“Food products,” “Non-food products,” and “Services”) and by 78 product groups. Such an approach makes it possible to identify hidden interrelations and specific factors affecting price dynamics in various segments of the consumer basket, taking into account regional characteristics. Calculations with monthly breakdowns were carried out on the basis of a wide range of methods: traditional econometric models (SARIMA, Prophet, and Ridge regression) and modern machine learning algorithms (Random Forest and CatBoost). The authors applied the principal component method (PCA) and recursive feature elimination (RFE) to improve the accuracy of the predicted data. Accuracy was assessed on the basis of cross-validation. The statistical significance of differences between models was checked using the Diebold-Mariano test. The results revealed that a disaggregated prediction approach provides higher accuracy compared to aggregated models. Particularly noticeable improvements are observed for commodity categories with high price volatility. The findings confirm that detailing the CPI structure in forecasting allows not only to increase the accuracy of estimates but also to obtain a more reliable analytical basis for making economic decisions in monetary policy. Authors’ contribution. Yu.A. Metel – development of forecasting methodology, model implementation, calculation, and sampling methods; N.N. Kunitsyna – literature review, description of results, drawing conclusions, and editing article text.