ARIMA-Profile of the Inflation Forecasting in the Tomsk Region Problem

Abstract

The inflation forecasting task is one of the key tasks for the Central Bank. Today it especially needs new approaches designed not only to serve as a tool for future inflation values predicting but also to expand knowledge about the structure of the process under study. Traditionally the Central Bank uses ARIMA time series models as a tool for infl ation forecasting. In this paper the concept of an ARIMA-profi le is introduced for the fi rst time as a set of suitable for solving a specifi c problem ARIMA-models general characteristics. Experimental confirmation of the ARIMA-profi le existence for the Tomsk region Consumer Price Index forecasting task is presented. To do this a series of experiments was conducted in each of which the models were ranked according to their predictive power. Then using heat maps and probability density diagram it was shown that all models can be divided into two groups — "suitable" and "not suitable" for the inflation forecasting problem. The ARIMA-profile was presented as a set of rules separating the fi rst group of models from the second. Such a set of rules was obtained as a result of solving the problem of classifying models into "suitable" and "not suitable" by the method of decision trees. The main template of models suitable for forecasting inflation in the Tomsk region has the ARIMA(*,*,*)(2,*,1)12 scheme. A common feature of the "best" models is the non-zero seasonal parameters P and Q and the structure of the series is such that the current inflation value is best described by the value of inflation which was observed a year ago and two years ago as well as the value of last year’s shock.