Inflation Forecasting: The Practice of Using Synthetic Procedures

Abstract

The article contains a review of inflation forecasting models, including the most popular class of models as one-factor models: random walk, direct autoregression, recursive autoregression, stochastic volatility with an unobserved component and of the integrated model of autoregression with moving average. Also, we discussed the possibilities of various modifications of models based on the Phillips curve (including the “triangle model”), vector autoregressive models (including the factor-extended model of B. Bernanke’s vector autoregression), dynamic general equilibrium models and neural networks. Further, we considered the comparative advantages of these classes of models. In particular, we revealed a new trend in inflation forecasting, which consists of the introduction of synthetic procedures for private forecasts accounting obtained by different models. An important conclusion of the study is the superiority of expert assessments in comparison with all available models. We have shown that in the conditions of a large number of alternative methods of inflation modelling, the choice of the adequate approach in specific conditions (for example, for the Russian economy of the current period) is a non-trivial procedure. Based on this conclusion, the authors substantiate the thesis that large prognostic possibilities are inherent in the mixed strategies of using different methodological approaches, when implementing different modelling tools at different stages of modelling, in particular, the multifactorial econometric model and the artificial neural network.